Stuck-busy alert has no known-wedged latch: re-fires every 6h forever with no escalate-once path

check_stuck_busy() (v4.66) re-alerts every STUCK_BUSY_REALERT_S=21600 forever while a wedge persists. There is no 'known-wedged, escalated once, stop repeating' state, so a wedge nobody clears becomes permanent 6-hourly noise on Elazar's DMs - and noise is how a correct alert gets ignored. Same gap CCM_REAP_EXCLUDE filled on the reap path for whey's parked proxy-* agents. Raised by nw-whey-cc 2026-08-09. Fix direction: escalate once loudly, then latch to silence (per-agent, cleared when the agent next reads idle), with a periodic rollup instead of per-agent repeats. Decide whether the latch belongs in agent_state or is derivable from actuator_log.