[P0] Cross-platform Docker matrix and CI workflows for hardware regression and automated tests

[Flink-ddd]

Context:
We need robust, reproducible build and test environments across our supported hardware backends to ensure code stability and catch hardware-specific regressions early.

Requirements & Exit Criteria:

• Set up Dockerfiles for CUDA and ROCm environments with necessary PyTorch and compiler dependencies.
• Configure GitHub Actions to run automated testing matrices across different hardware configurations.
• Ensure clear fallback behaviors are tested when hardware-specific dependencies (e.g., SM90 TMA) are missing.
• Document the Docker setup in our onboarding guide for source-build contributors.

[haoruilee]

@Flink-ddd I can take this one if it is still available 🙋🏻‍♀️

[Flink-ddd]

sure, but this task need NVIDIA and AMD GPUS, so Do you meet these basic hardware requirements?

[haoruilee]

Yes, I have access to both NVIDIA CUDA and AMD ROCm machines, so I can validate the Docker images and hardware CI paths.

[Flink-ddd]

That's great! we need set up a Cross-platform Docker matrix and CI workflows, so Please try it.