[Bugfix] Fix assertion error in flashmla backend with fullgraph enabled

[Kurumi5210]

Purpose

Fix an assertion error when using flashmla backend with fullgraph enabled.

Previously, _build_attention_metadata was called with num_tokens=num_tokens_unpadded. When pad_attn=True (required by fullgraph + flashmla), this leads to an inconsistency between num_tokens and padded attention-related metadata, triggering an assertion failure inside the attention backend.
This PR fixes the issue by passing num_tokens_padded when full cudagraph is enabled, ensuring consistency between num_tokens
Related to #33384

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 addresses an assertion error that occurs when using the flashmla backend with fullgraph enabled. The issue stemmed from an inconsistency in _dummy_run where _build_attention_metadata was called with an unpadded token count (num_tokens_unpadded) while other parameters were padded when pad_attn was true. This mismatch in padding led to the failure.

The proposed change corrects this by conditionally passing num_tokens_padded when pad_attn is true, ensuring all arguments to _build_attention_metadata are consistently padded. This is a direct and effective fix for the bug. The change is correct and well-contained.

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors.

You ask your reviewers to trigger select CI tests on top of fastcheck CI.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can either: Add ready label to the PR or enable auto-merge.

If you have any questions, please reach out to us on Slack at https://slack.vllm.ai.

🚀

[Kurumi5210]

@youkaichao @hmellor @robertgshaw2-redhat