Controller starts fine after upgrade but CRD drift makes new objects fail validation on only one cluster

The controller image upgrade succeeds, but one cluster still rejects new objects because its CRD version or schema was not updated along with the controller.