1) worker stopped: QueueLagGrowing firing after 60s; oldest job age=58s depth=57 jobs/s=0.00 worker_busy=0.0
   diagnosis from stored data: trace 4710543929ec0f33fc13ce8565b5bc6a has the api server and "attachments.jobs publish" spans and no consumer span; no "job ready" worker log in Loki during the last minute (0 lines)
   restored: worker started, alert resolved after 20s, backlog drained in 21s
PASS: stopped worker detected by alert and diagnosed from telemetry
