[aoti-et] Add a voxtral runner and add CI

[larryliu0820]

This pull request introduces an end-to-end CUDA test for the Voxtral model, adds a new runtime executable for Voxtral, and makes supporting updates to the build system and utility code. The main focus is on enabling automated validation of Voxtral's CUDA export and runtime within CI, including latency measurement and output verification.

End-to-end Voxtral CUDA test integration:

• Added a new test-voxtral-cuda-e2e job to the .github/workflows/cuda.yml CI workflow, which builds, exports, and runs the Voxtral model using CUDA, and checks for expected output and exit codes.
• Updated the optimum-executorch commit pin in .ci/docker/ci_commit_pins/optimum-executorch.txt to ensure compatibility with the latest Voxtral export.

Voxtral runtime and build system enhancements:

• Added a new voxtral_runner executable to backends/cuda/CMakeLists.txt for running exported Voxtral models, linking it with required CUDA and extension libraries.
• Introduced the implementation of voxtral_runner.cpp, which loads the model, runs the main methods (audio_encoder, token_embedding, text_decoder), prints tensor summaries, and reports method and run latencies.

Utility and compatibility updates:

• Updated dtype_to_scalar_type in backends/aoti/utils.h to support PyTorch's int64 dtype code, improving tensor type handling for Voxtral inputs.

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/14875

• 📄 Preview Python docs built from this PR

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

❗ 1 Active SEVs

There are 1 currently active SEVs. If your PR is affected, please view them below:

• Driver update on H100 and A100 instances

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