perf: convert 1 Diffusion allocation + add Memory imports to 17 layer files

Minor additions to the batch migration.

Running totals for #1014:
- NN Layers: 1498 raw + 332 pooled (was 1722 raw + ~100 pooled)
- RWKV7Block: 20 hot-path allocations → LayerWorkspace (232x speedup)
- All layers: 225 per-forward allocations → TensorAllocator.Rent

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: ooples

src/Diffusion/Audio/ShortTimeFourierTransform.cs

@@ -1,4 +1,4 @@
-using AiDotNet.Interfaces;
+using AiDotNet.Interfaces;
 using AiDotNet.LinearAlgebra;
 using AiDotNet.Tensors.Engines;
 using AiDotNet.WindowFunctions;
@@ -420,7 +420,7 @@ private Tensor<T> InverseBatched(Tensor<Complex<T>> spectrograms, int? targetLen
         }
         outputLength = targetLength ?? outputLength;
 
-        var output = new Tensor<T>(new[] { batchSize, outputLength });
+        var output = TensorAllocator.Rent<T>(new[] { batchSize, outputLength });
 
         for (int b = 0; b < batchSize; b++)
         {