Preserve all architectures in base image mirrors

[robobryce]

Summary

• mirror complete OCI image indexes with skopeo copy --all
• compare and verify source and target platform sets, including single-platform images
• run the mirror immediately when this workflow changes on any branch
• retain the existing PyTorch 2.12.1 mirror entry

Why

The CUDA 13.1 source currently publishes linux/amd64 and linux/arm64, but the GHCR mirror contains only linux/amd64. PR #213 builds the final PyHPC event image natively on both architectures and therefore requires the ARM64 base.

Merge this PR first and wait for the main push workflow to repair the mirror before merging or retriggering #213.

Validation

• actionlint
• yamllint with the repository configuration
• git diff --check
• live platform inspection of all three sources and GHCR targets
• fork GitHub Actions copy run published an index with the exact source digest and both architectures
• automatic all-branch push run verified all three mirror matrix entries

[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.

[github-actions]

👋 Thank you for your contribution!

This pull request is from a forked repository so GitHub Actions will not be able to run CI. A maintainer will review your changes shortly and manually trigger the CI.

@maintainers Please review this PR when you have a chance and follow the instructions in the CONTRIBUTING.md file to trigger the CI.

[robobryce]

Ready for maintainer vetting. Required order: merge this PR first, then wait for the main push run of Mirror Base Images to GHCR and verify the CUDA target contains both linux/amd64 and linux/arm64. The current target is AMD64-only.

[robobryce]

End-to-end fork validation is green. Copy run 29827736041 used the workflow token to publish a fresh CUDA package; its source and GHCR target have the identical index digest sha256:eaded8d05ec577d25b882256d99da4270caf4787ec5c79408b0c1286c1efc0f1 and both linux/amd64 and linux/arm64. Idempotence run 29828241988 then detected the matching platform set, skipped copying, and passed verification.

[brycelelbach]

Why restrict this to main? Don't we want it to run for all branches?

[robobryce]

Agreed. Removed the branch restriction in d20deae; the path filter still limits push runs to workflow-file changes, now on any branch. The automatic fork push run passed all three matrix entries: https://github.com/robobryce/accelerated-computing-hub/actions/runs/29829542820