Add torch_dma_kv_cache_benchmark.py under examples/microbenchmarks/ providing PyTorch parity with jax_dma_kv_cache_benchmark.py:

[copybara-service]

Add torch_dma_kv_cache_benchmark.py under examples/microbenchmarks/ providing PyTorch parity with jax_dma_kv_cache_benchmark.py:

• Single-host raw PCIe DMA microbenchmark testing KVCacheManager D2H and H2D throughput and latency across variable layers (1 to 128) and dtypes (bf16, fp32, int32, fp8).
• Implements comparative baselines against native PyTorch pinned host memory (pin_memory=True, non_blocking=True) and standard unpinned host memory.
• Logs statistical metrics (mean, median, sample stddev, 95% confidence intervals, throughput in GB/s) to JSONL telemetry log for golden result comparisons in CI/cron.
• Registers :torch_dma_kv_cache_benchmark_test target in examples/microbenchmarks/BUILD.
• Updates examples/microbenchmarks/README.md with execution instructions.

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.