kubelet and kube-proxy hang due to systemd start limit

In a CI test run, we saw kube-proxy hang due to systemd start rate limiting.

The charm had just upgraded to k8s 1.26 snaps, which caused kube-proxy to crash since it was still using a deprecated --logtostderr argument. These crashes caused kube-proxy to hit the systemd start limit, where no further restarts will be attempted.

The charm then reconfigured the service and tried to restart it, but failed:

2022-12-09 23:34:40 INFO unit.kubernetes-worker/1.juju-log server.go:316 Restarting kubelet and kube-proxy.
2022-12-09 23:34:40 WARNING unit.kubernetes-worker/1.upgrade logger.go:60 Job for snap.kube-proxy.daemon.service failed because the control process exited with error code.
2022-12-09 23:34:40 WARNING unit.kubernetes-worker/1.upgrade logger.go:60 See "systemctl status snap.kube-proxy.daemon.service" and "journalctl -xe" for details.

Which failed due to:

Dec 9 23:34:40 juju-86e872-8 systemd[1]: snap.kube-proxy.daemon.service: Start request repeated too quickly.
Dec 9 23:34:40 juju-86e872-8 systemd[1]: snap.kube-proxy.daemon.service: Failed with result 'exit-code'.
Dec 9 23:34:40 juju-86e872-8 systemd[1]: Failed to start Service for snap application kube-proxy.daemon.

No further restart attempts occurred.