CI: make ATOM test workflow reusable

[gyohuangxin]

Summary

• split atom-test.yaml into a reusable workflow_call entrypoint plus a lightweight trigger wrapper
• parameterize checkout, repo/ref, fork handling, model scope, and dashboard publishing so other repos can call the same ATOM test pipeline
• keep the existing push / pull_request / schedule / workflow_dispatch entrypoints via atom-test-trigger.yaml

Test plan

• Checked workflow YAML with editor diagnostics (ReadLints)
• Trigger the new wrapper workflow in ROCm/ATOM
• Invoke the reusable workflow from an external test workflow
• Verify pre-checkin signal lookup still works with the passed ATOM commit SHA
• Verify dashboard publishing still only runs from the wrapper on main push/schedule

[Copilot]

Pull request overview

This PR refactors the existing ATOM CI test pipeline into a reusable workflow_call workflow and introduces a lightweight trigger wrapper to preserve the current push / pull_request / schedule / workflow_dispatch entrypoints while enabling cross-repo reuse.

Changes:

• Converted .github/workflows/atom-test.yaml into a reusable workflow with workflow_call inputs/secrets for checkout source, fork handling, model scope selection, pre-check gating, and dashboard publishing.
• Added .github/workflows/atom-test-trigger.yaml as the wrapper workflow that maps repository events to the reusable workflow inputs and secrets.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
.github/workflows/atom-test.yaml	Reworked into a reusable workflow with parameterized checkout/source SHA, optional pre-check gating, and optional dashboard publishing.
.github/workflows/atom-test-trigger.yaml	New wrapper workflow that preserves existing triggers and calls the reusable workflow with computed inputs.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pre-checkin signal artifacts are published as checks-signal-${{ github.sha }} by pre-checks.yaml, i.e. keyed to the workflow run SHA (merge SHA for PRs). This job now sets GITHUB_SHA to env.GITHUB_COMMIT_SHA (which resolves to the PR head SHA via source_commit_sha), so check_signal.sh will look for a non-existent artifact on PRs and always time out. Use the checked-out ref/SHA (inputs.checkout_ref, which the trigger sets to github.sha) for the signal lookup, or introduce a dedicated signal_commit_sha input to avoid conflating the two SHAs.

Suggested change

	GITHUB_SHA: ${{ env.GITHUB_COMMIT_SHA }}
	GITHUB_SHA: ${{ inputs.checkout_ref }}

[Copilot]

publish_dashboard can cause this job to push to gh-pages, but accuracy-dashboard does not declare permissions: contents: write (and likely actions: read for artifact download). In reusable workflows, callers often set the default GITHUB_TOKEN permissions to read-only, which will make the dashboard publish fail even when publish_dashboard is true. Add an explicit permissions block on this job (or document/require the caller to grant these permissions) so reuse is reliable.

Suggested change

	runs-on: ubuntu-latest
	runs-on: ubuntu-latest
	permissions:
	contents: write
	actions: read

[Copilot]

The new run_precheck / source_commit_sha inputs allow callers to enable the pre-checkin gate, but GITHUB_COMMIT_SHA currently falls back to inputs.checkout_ref (which may be a branch/ref string). Since the pre-checkin artifact naming is SHA-based (checks-signal-<sha>), enabling run_precheck with a non-SHA checkout_ref will deterministically fail after retries. Consider validating that the commit input is a full SHA (or resolve the checked-out ref to a SHA via git rev-parse) before running the signal download to fail fast with a clear error.