Need to better define where stale config exceptions may happen, and catch them appropriately. In particular, we probably do not need to throw SCEs on the mongos side unless recv'd from mongod, if we detect a stale config on a mongos node we should simply fix and continue.