[Feat] upgrade FLA to v0.5.0

[KevinZeng08]

📌 Description

• upgrade FLA to v0.5.0 to support GVA feature, and apply some FLA bug fixes
• FLA v0.5.0 has a bug with cpu overhead which decreases perf of small sequences (https://github.com/fla-org/flash-linear-attention/blob/main/fla/ops/backends/__init__.py#L71), fixed in [Fix] Add functools.cache for op dispatch to reduce CPU overhead fla-org/flash-linear-attention#881, and all benchmarks are run with this fix

🔍 Related Issues

#67 #55

🚀 Pull Request Checklist

Thank you for contributing to cuLA! Before we review your pull request, please make sure the following items are complete.

✅ Pre-commit Checks

• I have installed pre-commit by running pip install pre-commit (or used your preferred method).
• I have installed the hooks with pre-commit install.
• I have run the hooks manually with pre-commit run --all-files and fixed any reported issues.

If you are unsure about how to set up pre-commit, see the pre-commit documentation.

🧪 Tests

• Tests have been added or updated as needed.
• All tests are passing.

⚡ Performance

Reviewer Notes

[gemini-code-assist]

Code Review

This pull request updates the flash-linear-attention (FLA) baseline to version 0.5.0 across benchmarks, documentation, and tests. It includes updated performance results for Hopper and Blackwell architectures and adjusts test cases to accommodate API changes in the FLA library. Feedback points out a stale reference to GB300 in the README and an inconsistency in the default output filename within the benchmark generation script.

[KevinZeng08]

Hi @sjmshsh , if you have time, you can help review this PR. Thanks~

[sjmshsh]

Hi @sjmshsh , if you have time, you can help review this PR. Thanks~

OK

[icavan]

LGTM

[zheyang0825]

LGTM

[zheyang0825]

LGTM