Allow executors to complete rather than terminate them

[ash211]

You can't read logs out of deleted pods but you can read logs out of completed pods. Completed pods are also reaped when their count exceeds a threshold so it's ok to have large numbers of completed pods -- we don't need to worry about deleting them to clear up disk space from logs (untested but believed to be standard k8s behavior).

This issue is to change the way to clean up executor pods from deleting them to instead having them complete (the JVM process exits cleanly) to preserve the logs.

Logs are accessible with: kubectl logs <pod-name>