Add video-scaling function for SFT padding

[hengtaoguo]

Description

Scale-to-fit before padding in video preprocessor and fix audio-in-video sequence serialization. This PR serves as the groundwork for varied-size video SFT:

• Scale-to-fit before padding: Improved smart_resize() preprocessing step in Qwen3-Omni image preprocessor to proportionally downscale videos exceeding max_h/w_pixels (resolving downstream shape crashes during batch padding) while preserving aspect ratio as closely as possible..
• Unit test: Added test_scale_to_fit_video_before_padding to verify correct aspect-ratio preserving downscaling and successful padding setup.
• Fixed a typo in add_extra_tokens_for_qwen3_omni where <|audio_pad|> (151675) was appended at the end of the interleaved video-audio sequence instead of <|audio_end|> (151670).

Tests

Example: Video of (220, 896) is resized to (128, 512) by new smart_resize: divisible by factor 32 and didn't exceed max canvas 512, preserving the aspect ratio as closely as possible:

python -m pytest tests/unit/qwen3_omni_layers_test.py::TestQwen3OmniMoeVisionPatchEmbed::test_scale_to_fit_video_before_padding -vv -s

Checklist

Before submitting this PR, please make sure (put X in square brackets):

• I have performed a self-review of my code. For an optional AI review, add the gemini-review label.
• I have necessary comments in my code, particularly in hard-to-understand areas.
• I have run end-to-end tests tests and provided workload links above if applicable.
• I have made or will make corresponding changes to the doc if needed, including adding new documentation pages to the relevant Table of Contents (toctree directive) as explained in our documentation.

[codecov]

Codecov Report

❌ Patch coverage is 0% with 11 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/maxtext/multimodal/processor_qwen3_omni.py	0.00%	11 Missing ⚠️

📢 Thoughts on this report? Let us know!

[github-actions]

🤖 Hi @hengtaoguo, I've received your request, and I'm working on it now! You can track my progress in the logs for more details.

[github-actions]

## 📋 Review Summary

This PR introduces video rescaling functionality (scale_to_fit_video_grid) to proactively prevent sequence shape crashes during padding by enforcing the video_max_grid_h/w constraints. The test additions are well thought out and the fix for audio_end token appending is correct.

🔍 General Feedback

• The core implementation of aspect-ratio preserving downscaling is solid and functionally sound.
• There are a few edge cases related to floating point precision and consistency of the image scaling factor between the intermediate and final steps of the preprocessing pipeline that should be addressed.
• Great job adding the unit tests to explicitly verify the grid limits and downscaling behavior!

[aireenmei]

Thanks for the work!