fix(ci): centralize GPU CI runtime pins

[yuanchen8911]

Summary

Centralizes the GPU CI runtime pins added in #694 behind the shared load-versions action. The GPU Operator chart version and snapshot-agent CUDA image remain in .settings.yaml, and the consuming scripts now receive those values from composite-action inputs/env instead of reading .settings.yaml directly.

Motivation / Context

This is a small follow-up to merged PR #694. In review, we moved the GPU Operator chart version and snapshot-agent CUDA image into .settings.yaml; this PR finishes that cleanup by making those pins follow the same load-versions path as other CI tool/image pins.

This keeps .settings.yaml as the source and .github/actions/load-versions as the shared read path.

Fixes: N/A
Related: #694

Type of Change

• Bug fix (non-breaking change that fixes an issue)
• New feature (non-breaking change that adds functionality)
• Breaking change (fix or feature that would cause existing functionality to change)
• Documentation update
• Refactoring (no functional changes)
• Build/CI/tooling

Component(s) Affected

• CLI (cmd/aicr, pkg/cli)
• API server (cmd/aicrd, pkg/api, pkg/server)
• Recipe engine / data (pkg/recipe)
• Bundlers (pkg/bundler, pkg/component/*)
• Collectors / snapshotter (pkg/collector, pkg/snapshotter)
• Validator (pkg/validator)
• Core libraries (pkg/errors, pkg/k8s)
• Docs/examples (docs/, examples/)
• Other: GPU CI composite actions and .settings.yaml

Implementation Notes

• Adds gpu_operator_chart_version and snapshot_agent_cuda_image outputs to .github/actions/load-versions.
• Updates runtime-install to load the GPU Operator chart version through load-versions for Helm-mode installs and pass it to install-gpu-operator-helm.sh via GPU_OPERATOR_CHART_VERSION.
• Updates aicr-build to load the snapshot-agent CUDA image through load-versions when build_snapshot_agent=true and pass it to build-snapshot-agent.sh via SNAPSHOT_AGENT_CUDA_IMAGE.
• Keeps optional action inputs for both values so future callers can override them explicitly if needed.
• Leaves the actual pinned values unchanged.

Testing

bash -n .github/actions/runtime-install/install-gpu-operator-helm.sh .github/actions/aicr-build/build-snapshot-agent.sh
yamllint .settings.yaml .github/actions/load-versions/action.yml .github/actions/runtime-install/action.yml .github/actions/aicr-build/action.yml
actionlint .github/workflows/gpu-smoke-test.yaml .github/workflows/gpu-h100-kind-runtime-test.yaml .github/workflows/gpu-h100-training-test.yaml .github/workflows/gpu-h100-inference-test.yaml
git diff --check

Scoped CI checks passed locally. Full make qualify was not run because this is a CI-only composite-action wiring change with no Go, recipe, or user-facing behavior changes.

Risk Assessment

• Low — Isolated change, well-tested, easy to revert
• Medium — Touches multiple components or has broader impact
• High — Breaking change, affects critical paths, or complex rollout

Rollout notes: Existing GPU CI callers keep using the same pinned values. The scripts now fail clearly if their parent action does not provide the required env value.

Checklist

• Tests pass locally (make test with -race)
• Linter passes (make lint)
• I did not skip/disable tests to make CI green
• I added/updated tests for new functionality
• I updated docs if user-facing behavior changed
• Changes follow existing patterns in the codebase
• Commits are cryptographically signed (git commit -S) — GPG signing info

[coderabbitai]

📝 Walkthrough

Walkthrough

This change refactors how configuration values are passed through GitHub Actions workflows. The load-versions action now extracts gpu_operator_chart_version and snapshot_agent_cuda_image from .settings.yaml and exposes them as outputs. The aicr-build and runtime-install actions accept these values as optional inputs and use conditional steps to load defaults when not provided. The corresponding shell scripts (build-snapshot-agent.sh and install-gpu-operator-helm.sh) are updated to validate that these values are set in environment variables rather than reading directly from .settings.yaml with yq. A new configuration entry is added to .settings.yaml for gpu_operator_chart_version.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~12 minutes

🚥 Pre-merge checks | ✅ 4

✅ Passed checks (4 passed)

Check name	Status	Explanation
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Description check	✅ Passed	The pull request description clearly explains the purpose of centralizing GPU CI runtime pins behind the load-versions action, provides motivation from a preceding PR, and details the implementation changes across multiple files.
Title check	✅ Passed	The title 'fix(ci): centralize GPU CI runtime pins' accurately reflects the main change: moving GPU-related configuration pins to a centralized location via the load-versions action.

_{✏️ 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.}

[mchmarny]

Clean follow-up to #694. Centralizes the two GPU CI pins through load-versions while keeping .settings.yaml as the source. The optional action inputs preserve override flexibility, the env-based pass-through is consistent with how other pinned values flow today, and the script-side guards correctly handle both empty and literal "null" from yq. End-to-end signal is strong: GPU smoke, H100 inference, and H100 training all green on this commit. LGTM.