1. Proposed title of this feature request
Properly clean unstaged volumes after node reboot

2. What is the nature and description of the request?
This is cloned from OCPBUGS-24324.

When the last pod that uses a CSI volume is deleted from a node and the node is rebooted before (or during) NodeUnstage call, a newly starter kubelet after the reboot will not call NodeUnstage and it will declare the volume unmounted (the volume is removed from node's status.volumesInUse).

This violates CSI spec, which says that NodeUnstage must be called before detaching the volume from the node, and it breaks some CSi drivers in a way that is very hard to fix (see OCPBUGS-24324). I'm tracking it as RFE, because it may need an upstream enhancement + feature gate.

3. Why does the customer need this? (List the business requirements here)

Customers need robust OCP that does not break their storage or volumes during node reboot.

4. List any affected packages or components.

Kubernetes / kubelet

This is already reported upstream as https://github.com/kubernetes/kubernetes/issues/121937