[Main2Main] Upgrade vllm commit to 0113

[wjunLu]

What this PR does / why we need it?

Upgrade vllm commit to 0113 (11b6af5)

• Modify import paths due to the refactors
  [1/N][Attention] Restructure attention: move files vllm#31916
  [3/N][Attention] Move AttentionMetadata-related code from utils.py to backend.py vllm#32054

• Fix TypeError: NPUOffloadingSpec.__init__() takes 2 positional arguments but 3 were given due to
  OffloadingConnector: Add cpu_bytes_to_use configuration vllm#24498

• Skip the async-scheduling tests in tests/e2e/multicard/4-cards/long_sequence/test_mtp.py, which are never verified
  [Misc] Enable async scheduling by default with spec decoding vllm#31998

• Skip some pooling tests, which are caused by
  [Model] Standardize pooling heads vllm#32148
  where vllm is also failed https://buildkite.com/vllm/ci/builds/46705/steps/canvas?jid=019bb329-3834-4685-862b-1613b8e0f5d4

  We will reopen those tests when main2main reachs [Bugfix] Replace PoolingParams.normalize with use_activation vllm#32243

• Skip some cases in tests/e2e/multicard/4-cards/long_sequence/test_mtp.py, which are broken by
  [Bugfix] Fix stale SSM state for new Mamba requests scheduled as decode vllm#32118

Does this PR introduce any user-facing change?

How was this patch tested?

• vLLM version: v0.13.0
• vLLM main: vllm-project/vllm@2f4e654

[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 the vllm dependency to a newer commit. The changes primarily consist of compatibility fixes to adapt to upstream API changes, which is a standard and well-executed approach for this type of update. The use of conditional imports and logic based on the vllm version is appropriate. I've identified one high-severity issue in a test file concerning the misuse of a context manager, which could lead to test pollution and flaky tests.

[gemini-code-assist]

set_current_vllm_config is a context manager and should be used with a with statement to ensure proper setup and teardown of the configuration context. Using it as a direct function call will cause the configuration to leak into other tests, potentially causing flakiness and hard-to-debug test failures. Since get_current_vllm_config is already being patched on line 217, this call is redundant and can be safely removed. The patch ensures that any code calling get_current_vllm_config will receive the mock config.

[wangxiyuan]

CI passed here: https://github.com/vllm-project/vllm-ascend/actions/runs/20992159837/job/60344387696?pr=5839

[wjunLu]

Conflicts solved.