[Core] [WIP] customize process between stages, remove the need of "request_ids" in model forward

[yinpeiqi]

PLEASE FILL IN THE PR DESCRIPTION HERE ENSURING ALL CHECKLIST ITEMS (AT THE BOTTOM) HAVE BEEN CONSIDERED.

Purpose

The related PR infos could be referred to [#92], the issue could be referred to [#74].

We add the custom process support for both qwen2.5omni model and the qwen3omni model.

Specifically, we allow user to custom a custom_preprocess function and custom_postprocess function between the forward function. In qwen3omni, we put the whole MTP stage into the custom_preprocess.

By this mean, in model_kwargs, we can remove request_ids, request_span, and addition_information_by_req.

Test Plan

Test Result

---

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan, such as providing test command.
• The test results, such as pasting the results comparison before and after, or e2e results
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.
• (Optional) Release notes update. If your change is user facing, please update the release notes draft.

BEFORE SUBMITTING, PLEASE READ https://github.com/vllm-project/vllm-omni/blob/main/CONTRIBUTING.md (anything written below this line will be removed by GitHub Actions)

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Guard preprocess kwargs when additional info is absent

When has_preprocess is true the runner now always calls self.model.preprocess(..., **req_infos), but req_infos is set via getattr(req_state, "additional_information_cpu", None) a few lines above. For new requests without any additional_information this value is None, so the **req_infos expansion raises TypeError: 'NoneType' object is not a mapping before the model forward ever runs. This breaks talker-stage execution for requests that do not provide extra metadata. Default the dict to {} before unpacking or skip the call when no info is available.

Useful? React with 👍 / 👎.