- DRA kubelet: Fixed the kubelet to also clean up `ResourceSlices` in some additional failure scenarios (driver was removed forcibly or crashed and did not restart). ([#132058](https://github.com/kubernetes/kubernetes/pull/132058), [@pohly](https://github.com/pohly)) [SIG Node and Testing]

stored in `/var/lib/kubelet/pod_status_manager_state` is now can always be read back after kubelet restart.
  
  Since the checkpoint format was changed to fix the issue, if you are using the feature `InPlacePodVerticalScaling`, please clean up the state file `/var/lib/kubelet/pod_status_manager_state` when upgrading the kubelet as failrue to do it will lead to incompatible state formats and kubelet's failure to start. ([#126620](https://github.com/kubernetes/kubernetes/pull/126620), ...