[Perf] Support Flashinfer RoPE+Quant+KV update kernel for trtllm_mha backend for GPT-OSS

[elvischenv]

Motivation

This PR is to support Flashinfer rope_quantize_fp8_append_paged_kv_cache kernel for trtllm_mha backend and enable it on GPT-OSS.

Depends on Flashinfer 0.6.0: #15551

Accuracy

PR

[{'eval_name': 'aime25', 'model_name': 'gpt-oss-120b-high_temp1.0_20251223_191831', 'metric': 0.9166666666666666}]

main

[{'eval_name': 'aime25', 'model_name': 'gpt-oss-120b-high_temp1.0_20251223_193331', 'metric': 0.9083333333333333}]

Perf (GPT-OSS-120b TP8 con8)

PR: 5.5% perf gain

Median TPOT (ms):                        2.91

main

Median TPOT (ms):                        3.07

Modifications

Accuracy Tests

Benchmarking and Profiling

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.
• Follow the SGLang code style guidance.
• Work with maintainers to merge your PR. See the PR Merge Process

[gemini-code-assist]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

[nvpohanh]

This can be reviewed together with #19451 . They are very similar, except that one is for trtllm_mha and one is for trtllm_mla

[Fridge003]

For the accuracy results, which model are you testing on?
Can you please post accuracy results for MTP, to make sure its acceptance length doesn't drop

[Fridge003]

Instead of adding this method in base class, can we control this fusion with an environ flag?
Now it is set to False by default. After this feature stabilizes it can be turned on by default