With shards authoritative we've replaced the filtering metadata recovery process with a new one that resides entirely on the shard durably.

The metadata itself is recovered on first access to the collection, however it would be desirable to have a background process that randomly clears out the in-memory state in order to exercise the recovery path more often than just once at startup.

Note that we do get some coverage of this with the background secondary restart hook since it will have the same effect as clearing the metadata, but this should exercise it a lot more often and with a lighter weight process.