Use maximum number of batched tokens to autotune MoE

[nvjullin]

Purpose

Follow up on #27904.
CC @varun-sundar-rabindranath.

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 in the Google Doc.

[gemini-code-assist]

Code Review

This pull request updates the Mixture-of-Experts (MoE) autotuning logic to use the maximum number of batched tokens from the scheduler configuration, which is a more appropriate parameter for this purpose than the CUDA graph capture size. The changes are logical, but the refactoring is incomplete, leading to a critical issue where a removed attribute is still being accessed in the code.

[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".