fix(neuralnetworks): restore Predict override for models with custom Forward

PR #1155 replaced `public override Tensor<T> Predict(Tensor<T> input) =>
Forward(input)` with `protected override Tensor<T> PredictEager(Tensor<T>
input) => Forward(input)` in 11 network classes. That reroutes Predict
through `NeuralNetworkBase.PredictCompiled` → `CompiledModelHost.Predict`
→ the compiled-replay plan.

The compiled-replay plan silently truncates the forward pass for the
affected models: on a ResNet-18 with 3×32×32 input the plan returns a
shape-[64] tensor (output of the first conv block's 64 channels)
instead of the expected shape-[10] logits. The net effect is wrong
output for every `Predict` call on these 11 model classes.

Verified with a per-call diagnostic:
  Forward direct shape:     [10]
  Predict compile-off shape: [10]
  Predict compile-on shape:  [64]   ← the regression

Root cause is in the tracer / compiled-replay machinery (the tracer
does not capture the shape-conditional control flow in Forward —
rank-3 → rank-4 promotion, final Reshape, etc.). That's a larger
infrastructure fix; this commit restores master's previous behavior so
`Predict` calls Forward directly, matching the pre-#1155 contract.

Affected models (all had master's `public override Predict` changed
to `protected override PredictEager` by #1155):
  - ResNetNetwork, VGGNetwork, EfficientNetNetwork, MobileNetV2Network,
    ConvolutionalNeuralNetwork, UNet3D, VoxelCNN
  - FastText, GloVe, Word2Vec, SiameseNeuralNetwork

Unblocks CI on every open PR that merges master (PR #1163, PR #1165).
Once the compiled-plan tracer is hardened to preserve shape-conditional
control flow, these overrides can be removed again.

Author: ooples

src/NeuralNetworks/ConvolutionalNeuralNetwork.cs

@@ -235,10 +235,13 @@ public override void UpdateParameters(Vector<T> parameters)
     /// </para>
     /// </remarks>
     /// <summary>
-    /// Routes inference through <see cref="NeuralNetworkBase{T}.PredictCompiled"/> for
-    /// compiled-plan replay; <see cref="Forward"/> remains the eager fallback.
+    /// Runs <see cref="Forward"/> directly rather than routing through the
+    /// compiled-replay path. <see cref="Forward"/> contains shape-conditional
+    /// control flow that the tracer does not capture faithfully — routing
+    /// through <see cref="NeuralNetworkBase{T}.PredictCompiled"/> was observed
+    /// to return an intermediate feature-map shape instead of the final logits.
     /// </summary>
-    protected override Tensor<T> PredictEager(Tensor<T> input) => Forward(input);
+    public override Tensor<T> Predict(Tensor<T> input) => Forward(input);
 
     /// <summary>
     /// Trains the convolutional neural network using the provided input and expected output.

src/NeuralNetworks/EfficientNetNetwork.cs

@@ -294,10 +294,13 @@ public Tensor<T> Forward(Tensor<T> input)
 
     /// <inheritdoc />
     /// <summary>
-    /// Routes inference through <see cref="NeuralNetworkBase{T}.PredictCompiled"/> for
-    /// compiled-plan replay; <see cref="Forward"/> remains the eager fallback.
+    /// Runs <see cref="Forward"/> directly rather than routing through the
+    /// compiled-replay path. <see cref="Forward"/> contains shape-conditional
+    /// control flow that the tracer does not capture faithfully — routing
+    /// through <see cref="NeuralNetworkBase{T}.PredictCompiled"/> was observed
+    /// to return an intermediate feature-map shape instead of the final logits.
     /// </summary>
-    protected override Tensor<T> PredictEager(Tensor<T> input) => Forward(input);
+    public override Tensor<T> Predict(Tensor<T> input) => Forward(input);
 
     /// <inheritdoc />
     public override void Train(Tensor<T> input, Tensor<T> expectedOutput)

src/NeuralNetworks/FastText.cs

@@ -245,10 +245,13 @@ public override void UpdateParameters(Vector<T> parameters)
         }
 
         /// <summary>
-        /// Routes inference through <see cref="NeuralNetworkBase{T}.PredictCompiled"/> for
-        /// compiled-plan replay; <see cref="Forward"/> remains the eager fallback.
+        /// Runs <see cref="Forward"/> directly rather than routing through the
+        /// compiled-replay path. <see cref="Forward"/> contains shape-conditional
+        /// control flow that the tracer does not capture faithfully — routing
+        /// through <see cref="NeuralNetworkBase{T}.PredictCompiled"/> was observed
+        /// to return an intermediate feature-map shape instead of the final output.
         /// </summary>
-        protected override Tensor<T> PredictEager(Tensor<T> input) => Forward(input);
+        public override Tensor<T> Predict(Tensor<T> input) => Forward(input);
 
         /// <summary>
         /// Trains the model on a single step of data using standard backpropagation.

src/NeuralNetworks/GloVe.cs

@@ -294,10 +294,13 @@ public override void UpdateParameters(Vector<T> parameters)
         /// returns their addresses (embeddings).
         /// </remarks>
         /// <summary>
-        /// Routes inference through <see cref="NeuralNetworkBase{T}.PredictCompiled"/> for
-        /// compiled-plan replay; <see cref="Forward"/> remains the eager fallback.
+        /// Runs <see cref="Forward"/> directly rather than routing through the
+        /// compiled-replay path. <see cref="Forward"/> contains shape-conditional
+        /// control flow that the tracer does not capture faithfully — routing
+        /// through <see cref="NeuralNetworkBase{T}.PredictCompiled"/> was observed
+        /// to return an intermediate shape instead of the final output.
         /// </summary>
-        protected override Tensor<T> PredictEager(Tensor<T> input) => Forward(input);
+        public override Tensor<T> Predict(Tensor<T> input) => Forward(input);
 
         /// <summary>
         /// Trains the model on a batch of word pairs and their co-occurrence counts.

src/NeuralNetworks/MobileNetV2Network.cs

@@ -289,10 +289,13 @@ public override Dictionary<string, Tensor<T>> GetNamedLayerActivations(Tensor<T>
 
     /// <inheritdoc />
     /// <summary>
-    /// Routes inference through <see cref="NeuralNetworkBase{T}.PredictCompiled"/> for
-    /// compiled-plan replay; <see cref="Forward"/> remains the eager fallback.
+    /// Runs <see cref="Forward"/> directly rather than routing through the
+    /// compiled-replay path. <see cref="Forward"/> contains shape-conditional
+    /// control flow that the tracer does not capture faithfully — routing
+    /// through <see cref="NeuralNetworkBase{T}.PredictCompiled"/> was observed
+    /// to return an intermediate feature-map shape instead of the final logits.
     /// </summary>
-    protected override Tensor<T> PredictEager(Tensor<T> input) => Forward(input);
+    public override Tensor<T> Predict(Tensor<T> input) => Forward(input);
 
     /// <inheritdoc />
     public override void Train(Tensor<T> input, Tensor<T> expectedOutput)

src/NeuralNetworks/ResNetNetwork.cs

@@ -493,17 +493,18 @@ public override void UpdateParameters(Vector<T> parameters)
     }
 
     /// <summary>
-    /// Makes a prediction using the ResNet network for the given input.
+    /// Makes a prediction using the ResNet network for the given input. Runs
+    /// <see cref="Forward"/> directly rather than routing through the
+    /// compiled-replay path, because <see cref="Forward"/> contains
+    /// shape-conditional control flow (rank-3 → rank-4 batch promotion and the
+    /// final <c>Reshape</c> that strips the synthetic batch dim) that the
+    /// tracer does not capture faithfully — routing through
+    /// <see cref="NeuralNetworkBase{T}.PredictCompiled"/> was observed to
+    /// return an intermediate feature-map shape instead of the final logits.
     /// </summary>
     /// <param name="input">The input tensor to make a prediction for.</param>
     /// <returns>The predicted output tensor containing class probabilities.</returns>
-    /// <summary>
-    /// Routes inference through <see cref="NeuralNetworkBase{T}.PredictCompiled"/> so the
-    /// forward pass gets traced and replayed by <c>CompiledModelHost</c> after warmup —
-    /// matching PyTorch's <c>torch.compile</c> default. The eager forward is
-    /// <see cref="Forward"/>, which retains the GPU-resident optimization path.
-    /// </summary>
-    protected override Tensor<T> PredictEager(Tensor<T> input) => Forward(input);
+    public override Tensor<T> Predict(Tensor<T> input) => Forward(input);
 
     /// <summary>
     /// Trains the ResNet network using the provided input and expected output.

src/NeuralNetworks/SiameseNeuralNetwork.cs

@@ -302,10 +302,13 @@ public override void UpdateParameters(Vector<T> parameters)
         }
 
         /// <summary>
-        /// Routes inference through <see cref="NeuralNetworkBase{T}.PredictCompiled"/> for
-        /// compiled-plan replay; <see cref="Forward"/> remains the eager fallback.
+        /// Runs <see cref="Forward"/> directly rather than routing through the
+        /// compiled-replay path. <see cref="Forward"/> contains shape-conditional
+        /// control flow that the tracer does not capture faithfully — routing
+        /// through <see cref="NeuralNetworkBase{T}.PredictCompiled"/> was observed
+        /// to return an intermediate shape instead of the final output.
         /// </summary>
-        protected override Tensor<T> PredictEager(Tensor<T> input) => Forward(input);
+        public override Tensor<T> Predict(Tensor<T> input) => Forward(input);
 
         /// <summary>
         /// Trains the model on pairs of inputs using a similarity learning objective.

src/NeuralNetworks/UNet3D.cs

@@ -237,10 +237,13 @@ public Tensor<T> Forward(Tensor<T> input)
     /// <returns>The predicted segmentation map.</returns>
     /// <inheritdoc />
     /// <summary>
-    /// Routes inference through <see cref="NeuralNetworkBase{T}.PredictCompiled"/> for
-    /// compiled-plan replay; <see cref="Forward"/> remains the eager fallback.
+    /// Runs <see cref="Forward"/> directly rather than routing through the
+    /// compiled-replay path. <see cref="Forward"/> contains shape-conditional
+    /// control flow that the tracer does not capture faithfully — routing
+    /// through <see cref="NeuralNetworkBase{T}.PredictCompiled"/> was observed
+    /// to return an intermediate feature-map shape instead of the final output.
     /// </summary>
-    protected override Tensor<T> PredictEager(Tensor<T> input) => Forward(input);
+    public override Tensor<T> Predict(Tensor<T> input) => Forward(input);
 
     /// <summary>
     /// Trains the network on a single batch of input-output pairs.

src/NeuralNetworks/VGGNetwork.cs

@@ -340,10 +340,14 @@ public override void UpdateParameters(Vector<T> parameters)
     /// </para>
     /// </remarks>
     /// <summary>
-    /// Routes inference through <see cref="NeuralNetworkBase{T}.PredictCompiled"/> for
-    /// compiled-plan replay; <see cref="Forward"/> remains the eager fallback.
+    /// Runs <see cref="Forward"/> directly rather than routing through the
+    /// compiled-replay path. <see cref="Forward"/> contains shape-conditional
+    /// control flow (batch-dim promotion, post-pipeline reshape) that the
+    /// tracer does not capture faithfully — routing through
+    /// <see cref="NeuralNetworkBase{T}.PredictCompiled"/> was observed to
+    /// return an intermediate feature-map shape instead of the final logits.
     /// </summary>
-    protected override Tensor<T> PredictEager(Tensor<T> input) => Forward(input);
+    public override Tensor<T> Predict(Tensor<T> input) => Forward(input);
 
     /// <summary>
     /// Trains the VGG network using the provided input and expected output.

src/NeuralNetworks/VoxelCNN.cs

@@ -230,10 +230,13 @@ public Tensor<T> Forward(Tensor<T> input)
     /// <returns>The predicted class probabilities or scores.</returns>
     /// <inheritdoc />
     /// <summary>
-    /// Routes inference through <see cref="NeuralNetworkBase{T}.PredictCompiled"/> for
-    /// compiled-plan replay; <see cref="Forward"/> remains the eager fallback.
+    /// Runs <see cref="Forward"/> directly rather than routing through the
+    /// compiled-replay path. <see cref="Forward"/> contains shape-conditional
+    /// control flow that the tracer does not capture faithfully — routing
+    /// through <see cref="NeuralNetworkBase{T}.PredictCompiled"/> was observed
+    /// to return an intermediate feature-map shape instead of the final output.
     /// </summary>
-    protected override Tensor<T> PredictEager(Tensor<T> input) => Forward(input);
+    public override Tensor<T> Predict(Tensor<T> input) => Forward(input);
 
     /// <summary>
     /// Trains the network on a single batch of input-output pairs.