[BUGFIX] Mtp torchair pd fix

[JC-ut0]

What this PR does / why we need it?

In memory of #2610
In the pd Disaggregation scenario, the first token of the inference after the d node receives the kv follows the eager mode.

Fixes:
Running with MTP torchair graph mode with Prefilling Decoding Disaggregation , if all requests processed by the D node are requests just transmitted from the P node, it will break the torchair graph.

Reason: During PD Disaggregation , the P node only transmits the KV cache and prompt to the D node, not the actual tokens inferred (neither the main model tokens nor the MTP tokens are transmitted). Therefore, the D node will treat this request as one without MTP tokens for inference (seq_len=1).
The community does not have graph mode issues because the community's attention has a seq_len=1 for each batch during the decode phase.
We have issues because the graph mode pads according to processing 2 tokens per request. When there are some seq_len=1 and some seq_len=2, padding is done at the end. If all requests received by the D node are seq_len=1, padding cannot be performed normally according to the attention's fia operator constraints.

Solution:

The kv consumer uses extra torchair graph padding to avoid breaking FIA graph constrains (The one this PR implemented).

The kv producer provides the correct tokens to the kv consumer, so that our graph mode constraints are not broken, and all logic is the same as the PD mixed deployment . Since we are using the community scheduler, the modification requires patching the vllm scheduler, but theoretically, performance should be better. (Maybe later )

Does this PR introduce any user-facing change?

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 a fix for an MTP torchair graph mode issue in prefilling-decoding disaggregation scenarios. The change involves adding extra padding to the torchair graph batch sizes for KV consumers to avoid breaking FIA graph constraints. However, the implementation has a logical flaw where self.torchair_graph_batch_sizes is modified and then reused, leading to incorrect calculations. I've suggested a refactoring to correct this logic.

[linfeng-yuan]

LGTM. Please check the logic of max_num_seq updating here.

Currently, deepseek can also performed with aclgraph in full graph mode, I think you can submit another PR to update the padding size of mtp for aclgraph also (full graph only).