upgrade vLLM to main

[wangxiyuan]

1. fix Update rope_scaling to rope_parameters in preparation for Transformers v5 vllm#28542
   The model structure modifications we involved in are:

   • Qwen2.5-VL(still exist some patch)
   • Qwen2-VL
   • Qwen2
   • DeepSeek series
   • Qwen-moe series

2. fix Revert "[Redo] #26368 (#28771)" vllm#29121
   the output token now type changed from np to list[list[int]]

3. fix [Core] Deprecate xformers vllm#29262
   xformers backend for multimodal now has been deprecated

4. fix [Attention] Remove imports from vllm/attention/__init__.py vllm#29342

5. fix [Core] Refactor padding logic and pad for CUDA graphs before attention metadata building  vllm#28579

6. fix [Feature] Prefill Context Parallel (PCP) basic support vllm#28718

7. fix [Config] Clean up SchedulerConfig initialization vllm#28665

8. fix [Frontend][torch.compile] CompilationConfig Overhaul (#20283): Set up -O infrastructure vllm#26847
   vllm introduced the optimization-level, some default config has been changed, and the param --enforce-eager has been deprecated

9. fix http://github.com/vllm-project/vllm/pull/29223 it retuns tuple for sampler.

10. fix Remove upstream fa checks vllm#29471 we'll remove the related patch to avoid this kind of error.

Co-authored-by: hfadzxy starmoon_zhang@163.com
Co-authored-by: wangli wangli858794774@gmail.com

• vLLM version: v0.11.2
• vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.2

[github-actions]

👋 Hi! Thank you for contributing to the vLLM Ascend project. The following points will speed up your PR merge:‌‌

• A PR should do only one thing, smaller PRs enable faster reviews.
• Every PR should include unit tests and end-to-end tests ‌to ensure it works and is not broken by other future PRs.
• Write the commit message by fulfilling the PR description to help reviewer and future developers understand.

If CI fails, you can run linting and testing checks locally according Contributing and Testing.

[gemini-code-assist]

Code Review

This pull request upgrades vLLM to main, which involves a lot of refactoring to align with upstream changes. Most changes are related to module path updates, API signature changes (e.g., rope_parameters), and data type changes (from numpy arrays to Python lists). I've found a critical issue in vllm_ascend/spec_decode/eagle_proposer.py where a hardcoded index is used instead of iterating through the batch, which will lead to incorrect behavior in speculative decoding. Please address this issue.

[github-actions]

This pull request has conflicts, please resolve those before we can evaluate the pull request.

[MengqingCao]

plz use english comment

[MengqingCao]

Plz fix the above comment and LGTM if CI passes