Recover logs from crashed and terminated pods

[sebasnallar]

Two layered fixes for the case where a pod's container has died but the log workflow returns nothing:

1. kube-logger-go now fetches the previous container instance's logs (PodLogOptions.Previous) on the first page for any pod whose application container has restarted or has a terminated last state. The kubelet retains exactly one prior instance per container, so this surfaces the crash output that kubectl logs alone hides while the replacement container is starting. Previous-container fetching is gated on the absence of a per-pod pagination cursor, so paginated requests stay unchanged: no extra round-trips, no duplicated lines (the existing timestamp-based dedup in processor handles overlap), and no token format change.

2. Deployment template flips terminationMessagePolicy from File to FallbackToLogsOnError on the application and traffic containers. When the file at /dev/termination-log is empty and the container exits non-zero, the kubelet auto-populates status.containerStatuses[].lastState.terminated.message with the tail (~80 kB) of stdout/stderr, which survives container restarts inside the Pod object itself. Apps that already write to the termination file are unaffected.

Together these cover CrashLoopBackOff and recently-terminated containers without any new in-cluster tooling. Pods that have been fully garbage-collected still require an external log aggregator.