Due to SERVER-50318, we only restart scheduled heartbeats. However, catchup expects all nodes to be restarted by resetting _updatedSinceRestart to false. Thus catchup may not wait for the latest heartbeats for all nodes and could exit catchup earlier. This is a regression.

On the restart for catchup. We should call hb.restart for all nodes rather than the scheduled. Alternatively, we could restart the heartbeats for all blindly on winning an election.