[long_seq_optim] BSND to TND and FA_UPDATE replacement

[pichangping]

What this PR does / why we need it?

We have optimized the performance of long sequences：First,Modify the input data format for attention calculation. Instead of using the original BSND format, remove the logic for converting between TND and BSND, and directly adopt the TND format.
The TND input format can be directly reused, which shortens the data flow path. Converting to BSND is an unnecessary processing step.Second, we switched the output update of the concatenated small operators to the npu_attention_update fusion operator to improve performance.

Does this PR introduce any user-facing change?

How was this patch tested?

• vLLM version: v0.11.0rc3
• vLLM main: vllm-project/vllm@c9461e0

[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 refactors the attention mechanism to leverage native TND layout support and a new npu_attention_update kernel on Ascend NPUs, removing the manual BSND packing/unpacking and update logic. While this simplifies the code and likely improves performance, I've identified several critical issues related to potential memory leaks in graph capture mode and a bug with a hardcoded value that could cause errors with models that have large context windows. There is also a high-severity concern about unnecessary type casting that could impact performance.

[zhenwenqi2024]

这里硬编码成2048,2048的原因是啥