[BugFix] Fix world size bug in model_runner

[wangxiyuan]

• Fix world size bug in model_runner to make sure ep>16 runs with MC2
• enable e2e test for vl

Co-Authored-By: whx-sjtu 2952154980@qq.com
Co-Authored-By: Icey 1790571317@qq.com

• vLLM version: v0.10.2
• vLLM main: vllm-project/vllm@3e903b6

[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.

[wangxiyuan]

rebase #2897 to test with the newest CI

[gemini-code-assist]

Code Review

This pull request addresses a bug in the selection of the MoE communication method within the model runner, specifically for scenarios involving expert parallelism on Ascend A2 hardware. The change correctly replaces world_size with world_size_across_dp in the condition to enable the 'mc2' communication method. This ensures that the decision is based on the world size relevant to expert parallelism, not the total world size which might include data parallelism dimensions. The accompanying test has been updated to reflect this logic change. The fix appears correct and is well-targeted to resolve the issue described.

[wangxiyuan]

also merged this one #2886