fix(worker): send SIGINT/SIGTERM to job process group on cancel/timeout
Fix SIGINT/SIGTERM not reaching the job process on cancel/timeout when FAVOR_UNSHARE_PID=true.
handle_child was sending SIGINT and SIGTERM only to the direct child PID (unshare), which ignores those signals, so the actual script never got them and always waited 5 seconds to be force-killed with SIGKILL. Sending the signal to the job's process group (-id) lets the script exit cleanly right away.
Summary by cubic
Fixes SIGINT/SIGTERM not reaching the job process on cancel/timeout when the job runs behind a non-signal-forwarding wrapper like unshare, tini, or sh. The worker now sends the signal to the process group recorded at spawn (a killpg) instead of only the direct child PID, so the script exits cleanly immediately rather than waiting to be force-killed with SIGKILL after 5 seconds. When process groups are disabled (DISABLE_PROCESS_GROUP), it still signals the direct child only. Failed signal deliveries are logged at debug level instead of panicking.