Capture abnormal runner death logs before the Pod is reaped

When a runner Pod dies abnormally mid-job (the GitHub ‘self-hosted runner lost communication’ case), the runners controller now reads the runner container’s final log tail and re-emits it to its own stdout before the Pod is deleted. This preserves the death trail in Loki even when the kubelet log would otherwise be garbage collected, so infrastructure deaths become diagnosable. The capture is gated on the runner’s exit code, so it records infrastructure deaths, not ordinary job failures.