Now, when preemption algorithm wants to delete victims, it only checks if the victim is waiting on permit - if it is, it's just rejected in that phase. But, if the victim is not there, the deletion API call is executed for such pod.

But then, this victim doesn't have to be bound yet, because it could wait in the PreBind phase for volume attachment or for some DRA stuff. Handling such scenario in better way could save us from making some API calls and won't require the victim to be recreated.

/sig scheduling
/cc @kubernetes/sig-scheduling-leads