Coordinator Start Command
How vigilo coordinator start runs configurable coordinator cycles continuously
until shutdown or service failure. Each cycle resumes a bounded number of
expired creating runs before chunk lease recovery and normal dispatch.
Dispatch claims one open run_id + run_shard cursor per window. Dispatch and
recovery insert one run.chunk.ready outbox event record per chunk, including
the chunk's run_shard, execution database_alias, and write_epoch in the
payload. Publication preserves that route hint so workers can claim
chunk-local work without an extra control-database routing lookup.