Pin openblas on aarch64

[jcrist]

We're seeing some test failures with the new openblas release on aarch64 (accuracy issues on comparison with sklearn's results, not issues in cuml's execution). I suspect a regression in openblas.

For now we pin openblas on aarch64.

[coderabbitai]

📝 Walkthrough

Walkthrough

The PR adds version constraints on libopenblas (upper bound ≤0.3.30) across CUDA-specific aarch64 environment configurations and introduces an architecture-specific dependency override in the manifest file to enforce this constraint for aarch64 targets.

Changes

Cohort / File(s)	Summary
Aarch64 CUDA Environment Files conda/environments/all_cuda-129_arch-aarch64.yaml, conda/environments/all_cuda-131_arch-aarch64.yaml	Added libopenblas<=0.3.30 dependency constraint to both CUDA 12.9 and 13.1 aarch64 environment configurations.
Dependency Manifest dependencies.yaml	Added architecture-specific output matrix override under dependencies.test_python.specific to enforce libopenblas<=0.3.30 for aarch64 targets in conda builds.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~8 minutes

🚥 Pre-merge checks | ✅ 3

✅ Passed checks (3 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title directly and specifically describes the main change: pinning openblas on aarch64 architecture, which matches the core modifications across all three modified files.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Description check	✅ Passed	The pull request description directly addresses the changeset by explaining test failures with openblas on aarch64 and pinning it to resolve accuracy issues.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[jameslamb]

I think this is ok for now to unblock CI, but we'll want to figure it out soon-ish. I could see a ceiling like this causing problems (especially in the all-of-RAPIDS environments like the devcontainers) as we look to add new Python versions and update other versions of things.

[jcrist]

Agreed that it's a stopgap. Once this is in I'll open a followup issue to investigate more. My brief observations are:

• 0.3.32 (just released): many many failures. Could possibly be fixed by an increase in tolerances, but there's enough failures here (and a suspicious upstream issue) that I suspect an openblas regression.
• 0.3.31 (released in jan, but for some reason wasn't used in our CI before???): only 2 small numeric failures, could be fixed by a slight increase in tolerances.

[jcrist]

/merge