Add Qwen3-30b-a3b RL recipe

[SurbhiJainUSC]

Description

This PR introduces a new RL training recipe for the Qwen3-30b-a3b model.

• Why is this change being made: To provide a standardized and optimized RL configuration for fine-tuning the Qwen3-30b-a3b architecture within MaxText.
• The problem being solved: Provides an out-of-the-box recipe for this model size and architecture, reducing the manual setup required for RL runs.
• Specific implementation: Adds the necessary configuration and script files to support this specific pipeline.

Tests

Tested on Ironwood cluster

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!

[hengtaoguo]

Should we still use XPK in the new recipes, considering its upcoming deprecation?