[WIP][Feature] Support MXFP8

[SlightwindSec]

This PR introduces support for MXFP8 (Microscaling Formats) data types for inference on the Ascend NPU (A5).

This PR is a continuation of the work originally started in PR #5113. Since that PR was closed, I have ported and adapted the codebase to the current master branch to ensure this feature reaches the community.

Huge thanks to @wangyao-i for the initial implementation and groundwork.

• vLLM version: v0.13.0
• vLLM main: vllm-project/vllm@7157596

[gemini-code-assist]

Code Review

This pull request introduces support for MXFP8 data types for inference on Ascend A5 NPUs. The changes are extensive, touching attention mechanisms, MoE communication, quantization configurations, and adding new quantization methods and parsers. Overall, the changes are well-structured to accommodate the new hardware capabilities.

I've identified a critical bug in the MoE token dispatcher that could lead to an UnboundLocalError, and a couple of high-severity issues related to code style and potential future bugs. Please address these points to ensure the stability and maintainability of the new features.

[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.

[github-actions]

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