[ROCm] Enable chunked prefill/paged attention in MLA on ROCm

[SageMoore]

This PR is largely just removing the guards in config.py to allow chunked prefill and paged attention in MLA. The LSE computation in the triton kernel doesn't work so we always fall back to flash attention in this case.

I ran lm_eval --model vllm --model_args pretrained=deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct,trust_remote_code=True,enable_chunked_prefill=True --tasks gsm8k --num_fewshot 5 --batch_size auto and got

vllm (pretrained=deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct,trust_remote_code=True,enable_chunked_prefill=True), gen_kwargs: (None), limit: None, num_fewshot: 5, batch_size: auto
|Tasks|Version|     Filter     |n-shot|  Metric   |   |Value |   |Stderr|
|-----|------:|----------------|-----:|-----------|---|-----:|---|-----:|
|gsm8k|      3|flexible-extract|     5|exact_match|↑  |0.7612|±  |0.0117|
|     |       |strict-match    |     5|exact_match|↑  |0.7437|±  |0.0120|

CC: @LucasWilkinson

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can either: Add ready label to the PR or enable auto-merge.

🚀

[vincent-4]

Thoughts?

[vincent-4]

Suggested change

	if is_hip and envs.VLLM_USE_TRITON_FLASH_ATTN and has_context is False:
	if is_hip and envs.VLLM_USE_TRITON_FLASH_ATTN and not has_context:

[vincent-4]

Suggested change

	has_context is False:
	not has_context: