[0.11.0][Perf] Add padding vision tower for Qwen2_5_Omni

[Semmer2]

Replace Qwen2_5_Omni_Thinker model's original vision tower Qwen2_5_VisionTransformer with AscendQwen2_5_VisionTransoformer. It pads attention qkv weights and bias for better performance.

What this PR does / why we need it?

This PR repalce the vision tower in Qwen2.5-Omni-Thinker model, Qwen2_5_VisionTransformer, with AscendQwen2_5_VisionTransformer, which use QKV padding for padding performance.

Does this PR introduce any user-facing change?

No

How was this patch tested?

• vLLM version: v0.11.0rc3
• vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.0

[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 introduces performance optimizations for the Qwen2.5-Omni-Thinker model on Ascend hardware by replacing the standard vision tower with an optimized version that uses QKV padding. The changes involve registering the new Ascend-specific model, implementing the model class which uses AscendQwen2_5_VisionTransformer, and updating the weight loading logic to support the new attn.qkv naming and apply padding.

My review has identified a critical issue in the new model's initialization that would cause a runtime error, and a high-severity issue regarding the robustness of the weight padding logic, which currently only handles one type of QKV layer naming. Please see the detailed comments for suggestions.

[MengqingCao]

I'm okay with merging this into 0.11.0-dev, lgtm