[webgpu] fix LayerNorm with empty input (microsoft#24244)

### Description
This PR fixes test case `CudaKernelTest.LayerNorm_NullInput`, in which
the input is 0-sized for LayerNorm.

`context.Output()` need to be called before returning.

Author: fs-eire

onnxruntime/core/providers/webgpu/nn/layer_norm.cc

@@ -74,10 +74,6 @@ Status LayerNorm<simplified>::ComputeInternal(onnxruntime::webgpu::ComputeContex
 
   const auto x_shape = x->Shape();
 
-  if (x_shape.Size() == 0) {
-    return Status::OK();
-  }
-
   const bool is_fp16 = x->GetElementType() == ONNX_TENSOR_ELEMENT_DATA_TYPE_FLOAT16;
 
   const size_t axis = NormalizeAxis(axis_, x_shape.NumDimensions());
@@ -109,6 +105,10 @@ Status LayerNorm<simplified>::ComputeInternal(onnxruntime::webgpu::ComputeContex
   auto* mean = context.Output(1, mean_shape);
   auto* inv_std_dev = context.Output(2, mean_shape);
 
+  if (x_shape.Size() == 0) {
+    return Status::OK();
+  }
+
   LayerNormProgram program{bias != nullptr, is_fp16, simplified, mean != nullptr, inv_std_dev != nullptr};
 
   program