Add Qwen3-Omni layers test to CI

[hengtaoguo]

Description

The transformers==5.13.0 Qwen3-Omni audio reference changed how it handles packed attention inputs, while MaxText still tests batched audio inputs directly.

This PR updates the Qwen3-Omni audio reference tests to match the current transformers API. Instead of flattening multiple batch items or chunks and passing an explicit block-diagonal attention_mask, the tests now run each independent batch item/chunk separately in the PyTorch reference path. This preserves the original no-cross-attention behavior without passing a duplicate attention_mask.

Also included:

• Skip the Qwen3-Omni preprocessing parity test, which depends on optional media packages such as decord.

Tests

python -m pytest tests/unit/qwen3_omni_layers_test.py -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

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!