refact attn metadata build

[weiguihua2]

What this PR does / why we need it?

1. Decouple attention metadata from the runner.
2. Use common input parameters for the attention metadata build and build_torchair_graph_dummy methods.
3. Use the community method split_decodes_and_prefills in mla to obtain data such as num_decodes.
4. MTP supports v1 schedule.

Does this PR introduce any user-facing change?

NO

How was this patch tested?

Environmental measurement

• vLLM version: v0.10.0
• vLLM main: vllm-project/vllm@4efd43e

[gemini-code-assist]

Code Review

This pull request refactors the attention metadata building process by introducing AscendCommonAttentionMetadata and TorchairCommonAttentionMetadata dataclasses. This is a positive change that improves code structure and decouples components from the main runner. However, the refactoring has introduced a few critical issues where the runner object is still being accessed after its removal from class initializers, which will lead to AttributeError exceptions. Additionally, there are some inconsistencies in the new dataclasses that will cause TypeError exceptions. I've provided detailed comments and suggestions to fix these issues.

[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.

[github-actions]

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

[github-actions]

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