add ragged gather reduce v2

[NuojCheng]

Description

Onboard the ragged_gather_reduce_v2 kernel to MaxText.

Performance Evaluation

Synthetic Data (Random Routing, EP=8)

• Before: xids/271201875 | xprof
• After: xids/271194893 | xprof

Observations:

• Step time improved, dropping from 14.24s to 13.17s.
• Unsort kernel time per layer dropped significantly from 13.97ms to 4.9ms.
• Sort kernel time remained unchanged.

Real Data (Without Random Routing, EP=4)

• Before: xids/271240396 | xprof
• After: xids/271240858 | xprof

Observations:

• Step time improved slightly, dropping from 17.8s to 17.5s.
• Unsort kernel time remained unchanged.

Tests

• Convergence: Tested in xids/271420711. The loss decreases normally.

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 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 91.75258% with 24 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
.../maxtext/kernels/ragged/ragged_gather_reduce_v2.py	91.72%	14 Missing and 10 partials ⚠️

📢 Thoughts on this report? Let us know!