Pod crash while HPA statefulsets scale down
The short answer is, if graceful termination fails, the pods are not restarts/rescheduled.
Pods may fail due to any reason (node issue, pod evicted etc.) in such scenario preStop hook is never invoked.
And this is by design, even if one writes a logic to restart/reschedule the pod will not help to restore the data from the pod (as it will be fresh pod).