Since we don't have a multi-version catalog, it's impossible to route changeStream cursors properly in a sharded cluster when the collection has been dropped. Therefore we will make it so that resuming change streams on a collection that has been dropped will error. To do this we will change the resumeToken to include the collection UUID instead of the namespace, and when a mongod gets a changeStream request with a resumeToken, it will first check if it has a collection with that UUID, and will error if it does not.