Implement coordinated restart of all pods on an affected NVMe subsystem

[noctarius]

Implementation for #298 (tracking issue). Depends on the investigation sub-issue.

Task

Implement the coordinated simultaneous restart chosen in the investigation: when a shared NVMe subsystem recovers, restart all pods owning all volumes on that subsystem together, so filesystem volumes restage without leaving any pod on a partially-torn-down subsystem.

Once this is in place, relax the intermediate workaround (skip-on-namespaced-volumes) where it is safe to do so, restoring auto-restage for namespaced volumes via the coordinated path.

Notes

• Exact mechanism and owning component (Guardian vs. operator) are decided by the investigation sub-issue.