[Bugfix] Disable the dispatch_ffn_combine kernel in MTP path

[kiscad]

What this PR does / why we need it?

This PR is to fix a smoking test failure. Adjust mtp_proposer and model_runner_v1 to route MTP decoding through the non‑fused MoE implementation while keeping the overall inference flow unchanged.

Does this PR introduce any user-facing change?

How was this patch tested?

This PR will be tested in smoking tests.

• vLLM version: v0.12.0
• vLLM main: vllm-project/vllm@ad32e3e

[gemini-code-assist]

Code Review

This pull request introduces a bugfix to disable the fused-moe kernel during the dummy_run of the MTP (Multi-path Transformer) proposer. This is accomplished by checking if the selected MoE communication method is FUSED_ALLTOALL and reverting to the standard ALLTOALL method if it is. This change is localized and specifically targets the dummy_run, which is crucial for graph capturing. The modification correctly addresses a likely bug with the fused kernel in this context, and the implementation is sound. No issues were found in the proposed changes.

[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.

[zzzzwwjj]

1. moe_comm_type judgement conditions need to consider whether is quant case;
2. dispatch_ffn_combine op need support EP>16 case;
   This two problems need to finish afterwards.