Improve CI investigation instructions to prevent agents from misreading workflow run status

[Copilot]

Agents were confusing completed/failure workflow runs with in_progress ones, causing them to browse stale logs and get confused by already-fixed failures.

Changes

• Anchor to the right run: instruct agents to first fetch the PR's head_sha via pull_request_read, then match it against list_workflow_runs results — avoiding accidental inspection of runs for older commits
• Verify live status: require calling get_workflow_run before reporting status; never rely on a previously-seen value
• Clarify status vs conclusion: explicitly document that status: "completed" + conclusion: "failure" is a done/failed run, not an in-progress one, and that conclusion must be checked separately from status
• Reorder steps to reflect the correct investigation flow: get head SHA → find matching run → verify status → list failed jobs → fetch logs

---

📱 Kick off Copilot coding agent tasks wherever you are with GitHub Mobile, available on iOS and Android.

[Simn]

I feel like we shouldn't need to teach GitHub Copilot how GitHub Actions works, but perhaps the existing instructions indeed confused it.