dispatch-suites

Dispatch and watch the heavy suites for branch $ARGUMENTS (default: the current branch). Reference: dev-docs/playbooks/ci-suites.md.

1. Check for in-flight runs first — re-dispatching a workflow on the same ref CANCELS its in-flight run (concurrency groups): gh run list --branch <branch> --limit 5
2. Dispatch:

   gh workflow run integration-tests-ci.yaml --ref <branch> -f test_threads=4
   gh workflow run test-cluster.yaml --ref <branch>
   gh workflow run ts-integration-tests.yaml --ref <branch>
   

3. Capture the three run IDs (gh run list --workflow <wf> --branch <branch> --limit 1) and arm a background watcher per run; report each conclusion as it lands rather than polling inline.
4. On a failure, download the artifact before triaging (rust-tests-log / cluster-tests-log-<attempt> / localnet-logs). Distinguish three failure shapes:
   • test failure (nextest exit 100 / vitest FAILED) → read the failing test's replay;
   • runner-pod death ("runner lost communication", no artifacts) → infra, re-dispatch once;
   • TS-suite wedge (cascading "Object does not exist" timeouts) → run dev-docs/playbooks/mpc-stall-postmortem.md on localnet-logs FIRST; the pre-existing epoch-entry race (issue #1736) must be ruled out before attributing the failure to the branch.