Trim unit test time

[RissyRan]

Description

Trim CPU & TPU unit tests time:

• In moe_test.py, update the test seq from default 2048 to 128
• In train compile test, update to a smaller topology if not specified in test name or restricted by sharding (v6e-256 --> v5p-128/64, etc)
• In train compile test, update to seq len to 1024, if not testing CP
• In train compile test, use DS v3-test version if not restricted to some configs (diff)
• Remove one duplicate test

Tests

• Expect tests are green
• CPU tests, saved: 7-8 min
• TPU tests saved: half time for MoE tests

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!

[NuojCheng]

Thank you Ran!