docs(agentic): Phase 3 — document quantization composition via ModelCompression (#1544)

Quantization for the local engine is achieved by quantizing the NeuralNetworkBase<T> with the existing ModelCompression stack before wrapping it — the adapter accepts any such network unchanged. No engine-side code (no duplication of ModelCompression). Documented on NeuralNetworkCausalLanguageModel.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Author: franklinic

src/Agentic/Models/Local/NeuralNetworkCausalLanguageModel.cs

@@ -19,6 +19,12 @@ namespace AiDotNet.Agentic.Models.Local;
 /// and calls <see cref="NeuralNetworkBase{T}.ResetState"/> before each pass so recurrent models (Mamba/GLA)
 /// start fresh — correct, if not yet optimal. A KV-cached fast path is a planned follow-up.
 /// </para>
+/// <para>
+/// <b>Quantization</b> composes through this adapter rather than being re-implemented here: quantize the
+/// network with the repository's <c>ModelCompression</c> stack first, then wrap the quantized
+/// <see cref="NeuralNetworkBase{T}"/> — the adapter accepts any such network unchanged, so a smaller/faster
+/// model needs no engine-side code.
+/// </para>
 /// <para><b>For Beginners:</b> This is the bridge that lets the local chat engine talk to a real AiDotNet
 /// network. It turns the running list of tokens into the exact tensor shape the network expects, asks the
 /// network for its prediction, and hands back the scores for the next token — which the engine then samples