Abnormal runner pod deaths are now recoverable in logs after reap

When a self-hosted runner pod dies mid-job, the controller now re-emits the runner container’s log tail to its own stdout before the pod is reaped. The captured trail (tagged with pod, pool, and endedAt) survives the Pod deletion, so abnormal infrastructure deaths like “self-hosted runner lost communication” are now diagnosable in Loki. Only abnormal exits are captured (non-zero exit, SIGKILL, or reap before clean exit), so failed job outputs are not re-logged.