Add CudaSampler class for GPU-based token sampling

[larryliu0820]

Add CudaSampler class that provides a high-level interface for GPU sampling:

• cuda_sampler.h: Class declaration with sample_argmax() method.
  Pre-allocates GPU memory to avoid allocation in hot path.
• cuda_sampler.cu: Implementation using the default CUDA stream (nullptr)
  for implicit synchronization with the CUDA backend's stream.

The default stream approach ensures proper ordering between decoder
output and argmax without requiring explicit cross-stream synchronization
or access to the backend's internal stream.

[larryliu0820]

Stack from ghstack (oldest at bottom):

• Integrate CUDA sampler into ASR runner and enable skip_copy for decoder #16388
• -> Add CudaSampler class for GPU-based token sampling #16387
• Add CUDA argmax kernel for LLM sampler #16386

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/16387

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 3 New Failures, 3 Unrelated Failures

As of commit 8573c9a with merge base c5d66a5 ():

NEW FAILURES - The following jobs have failed:

• Lint / lintrunner / linux-job (gh)
  >>> Lint for extension/llm/sampler/cuda_sampler.h:
• pull / test-multimodal-linux (gemma3-4b) / linux-job (gh)
  RuntimeError: Command docker exec -t d2dcb92572f560fa9697db3ec65a5397b8a1c4c6c3f2a132774c6ccdce50dc96 /exec failed with exit code 139
• Test Metal Backend / test-executorch-metal-build / macos-job (gh)
  RuntimeError: Command bash /Users/ec2-user/runner/_work/_temp/exec_script failed with exit code 1

BROKEN TRUNK - The following jobs failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

• pull / test-samsung-models-linux / linux-job (gh) (trunk failure)
  ##[error]The operation was canceled.
• pull / unittest / macos / macos-job (gh) (trunk failure)
  export/tests/test_target_recipes.py::TestTargetRecipes::test_mv3_model

UNSTABLE - The following job is marked as unstable, possibly due to flakiness on trunk:

• pull / android / run-emulator (gh) (#16137)
  Timeout waiting for emulator to boot.

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[Gasoonjia]

Im wondering if we really need to make sampler and cuda backend using same cuda stream, since the sampling and decoding should be able to work in parallel: the argmax process of logits_{i} should be able to work with the decoder generating logits_{i+1} since they do not have any dependency, and such parallelism may not happen if argmax and decoder share the same cudastream.