Improve Pytorch Copy Lock

[rich7420]

Purpose of PR

Now we make multiple copy in pytorch import like: torch.Tensor -> tolist() -> Vec.
This PR replaced the tolist() approach with zero-copy conversion using PyO3 NumPy interface:

• Convert PyTorch tensor to NumPy view via tensor.detach().numpy() (zero-copy when C-contiguous)
• Extract &[f64] slice directly using PyReadonlyArrayDyn::as_slice()
• Eliminates intermediate Python list and Rust Vec allocations

Otherwise, this pr provide a benchmark to test latency of pytorch.
We could remove the benchmark if you think it's unnecessary.

Related Issues or PRs

Changes Made

• Bug fix
• New feature
• Refactoring
• Documentation
• Test
• CI/CD pipeline
• Other

Breaking Changes

• Yes
• No

Checklist

• Added or updated unit tests for all changes
• Added or updated documentation for all changes
• Successfully built and ran all unit tests or manual tests locally
• PR title follows "MAHOUT-XXX: Brief Description" format (if related to an issue)
• Code follows ASF guidelines

[rich7420]

before

uv run python benchmark_latency_pytorch.py --qubits 16 --batches 100 --batch-size 32 2>&1
Uninstalled 1 package in 0.40ms
Installed 1 package in 2ms
PyTorch Tensor Encoding Benchmark: 16 Qubits, 3200 Samples
  Batch size   : 32
  Vector length: 65536
  Batches      : 100
  Prefetch     : 16

======================================================================
PYTORCH TENSOR LATENCY BENCHMARK: 16 Qubits, 3200 Samples
======================================================================

[Mahout-PyTorch] PyTorch Tensor Input (Zero-Copy Optimization)...
  Total Time: 2.8140 s (0.879 ms/vector)

[Mahout-NumPy] NumPy Array Input (Baseline)...
  Total Time: 2.3389 s (0.731 ms/vector)

======================================================================
LATENCY COMPARISON (Lower is Better)
Samples: 3200, Qubits: 16
======================================================================
PyTorch Tensor          0.879 ms/vector
NumPy Array             0.731 ms/vector
----------------------------------------------------------------------
Speedup: 0.83x
Improvement: -20.3%

[rich7420]

after

uv run python benchmark_latency_pytorch.py --qubits 16 --batches 100 --batch-size 32 2>&1
Uninstalled 1 package in 0.40ms
Installed 1 package in 4ms
PyTorch Tensor Encoding Benchmark: 16 Qubits, 3200 Samples
  Batch size   : 32
  Vector length: 65536
  Batches      : 100
  Prefetch     : 16

======================================================================
PYTORCH TENSOR LATENCY BENCHMARK: 16 Qubits, 3200 Samples
======================================================================

[Mahout-PyTorch] PyTorch Tensor Input (Zero-Copy Optimization)...
  Total Time: 2.3464 s (0.733 ms/vector)

[Mahout-NumPy] NumPy Array Input (Baseline)...
  Total Time: 2.3175 s (0.724 ms/vector)

======================================================================
LATENCY COMPARISON (Lower is Better)
Samples: 3200, Qubits: 16
======================================================================
PyTorch Tensor          0.733 ms/vector
NumPy Array             0.724 ms/vector
----------------------------------------------------------------------
Speedup: 0.99x
Improvement: -1.2%

[guan404ming]

LGTM

[guan404ming]

PyTorch Tensor Encoding Benchmark: 16 Qubits, 3200 Samples            
Batch size   : 32
Vector length: 65536
Batches      : 100
Prefetch     : 16

======================================================================
PYTORCH TENSOR LATENCY BENCHMARK: 16 Qubits, 3200 Samples
======================================================================

[Mahout-PyTorch] PyTorch Tensor Input (Zero-Copy Optimization)...
Total Time: 1.4448 s (0.451 ms/vector)

[Mahout-NumPy] NumPy Array Input (Baseline)...
Total Time: 1.4269 s (0.446 ms/vector)

======================================================================
LATENCY COMPARISON (Lower is Better)
Samples: 3200, Qubits: 16
======================================================================
PyTorch Tensor          0.451 ms/vector
NumPy Array             0.446 ms/vector
----------------------------------------------------------------------
Speedup: 0.99x
Improvement: -1.3%