Scheduled ingest jobs: a hung child import leaves a stale lock

In a scheduled agent job that ends by running a long CLI import (for example <tool> import and then an embed step), the session can end while the import keeps running. The job's lock file then stays behind. Later runs see a lock older than the timeout, replace it, and then skip the import again because a pgrep check finds the old process. Check the elapsed time and CPU of the matched process (ps -o etime,%cpu -p PID): hours of run time at about 3% CPU points to a hung import, not active work. Log the PID so a human can kill it. Run long imports with a timeout so a hang cannot block every later run.