Change CI to two fixed seed runs

[skyw]

We have been using two rounds of test in CI, one with fixed seed and one without. As we adding more complex tests, the error margin starts vary more with different seed. If we keep increasing threshold to allow maximum variance caused by random seed to pas, it can start gating real errors and defeat the purpose of having multiple rounds in CI.

The decision is to change to two rounds both with fixed seed. And more rounds can be added trivially by just adding more seed into the loop.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[skyw]

/ok to test 4c2847d

[greptile-apps]

Confidence Score: 5/5

[Medium risk] Changes test execution strategy from random to fixed seeds.

The PR appears safe to merge; the previously reported seed mismatch is fixed and no new actionable issue was established.

Findings

1.  Documented CI seeds differ ▶

Summary

The PR replaces the random L0 GPU test pass with a second fixed-seed pass and updates contributor guidance to name the CI seeds.

_{Reviews (2) · Last reviewed commit: "update docs"}

[skyw]

/ok to test b085453

[github-actions]

Test Results

   99 files  +  2    198 suites  +7   2m 10s ⏱️ -3s
1 464 tests ±  0  1 464 ✅ ±  0  0 💤 ±0  0 ❌ ±0 
3 507 runs  +163  3 507 ✅ +163  0 💤 ±0  0 ❌ ±0 

Results for commit b085453. ± Comparison against base commit 840ae70.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!