This is closely related to #8501. When updating most zones, the planner does an "expunge the old version" step then a "restore redundancy (by adding a new instance at the new version)" step. Currently it does the same thing for Nexus, which is incorrect: it will need to keep old Nexus instances around and also start up new Nexus instances concurrently, in order for the handoff work covered by #8501 to function.