Serialize Crane workflow runs

[mrjf]

fix(crane): serialize workflow runs

TL;DR

Crane runs can take longer than the schedule interval once a stale migration actually resumes work. This PR adds a repo-wide Crane concurrency group so scheduled and manual Crane invocations do not overlap and race the same migration state.

Problem

After the stale-completion scheduler fix, issue #78 was correctly selected as active work, but multiple Crane runs overlapped while earlier agent jobs were still inside Execute GitHub Copilot CLI. The compiled default top-level concurrency used github.run_id for schedule/workflow_dispatch events, which made each run its own lock instead of serializing Crane.

Approach

• Add source workflow concurrency to .github/workflows/crane.md:
  • group: crane-${{ github.repository }}
  • cancel-in-progress: false
• Run gh aw compile crane and include the regenerated .github/workflows/crane.lock.yml.

Validation

gh aw compile crane
✓ .github/workflows/crane.md (122.3 KB)
✓ Compiled 1 workflow(s): 0 error(s), 0 warning(s)

gh aw validate crane
✓ .github/workflows/crane.md
✓ Compiled 1 workflow(s): 0 error(s), 0 warning(s)

git diff --check

How to test

1. Dispatch Crane while another Crane run is active.
2. Confirm the second run waits under the shared crane-${{ github.repository }} concurrency group rather than starting another agent job.
3. Confirm scheduled runs no longer overlap long-running manual/scheduled Crane agents.

Trade-offs

• This serializes all Crane migrations in the repository, not just one migration. Crane already selects one migration per run, so repo-wide serialization matches the current operating model.
• cancel-in-progress: false preserves in-flight work instead of killing the active agent when a scheduled tick arrives.