Request connector run

Ask for an acquisition run now instead of at the next scheduled time, for example to check again a source that failed. The next run is pulled in, never pushed out, so repeating the request changes nothing and a run already in flight answers it. Rate limits still hold -- the run starts no sooner than the deployment interval floor (30 s by default) after the previous run ended, nor before the Retry-After the source asked for. The run then goes through the usual scheduler lease and records its outcome in health, committing connector.health_changed when the state changes; the request itself commits no event. run_at is when the run is due; the scheduler starts it within seconds after. A disabled instance is 409 connector_disabled; a paused instance is 409 connector_paused.