We have targeted test coverage for selective magic restore as part of SPM-2209, but we should extend our passthroughs to randomly perform selective restores. We'll get better coverage of different collection types (timeseries, capped collections, TTL indexes, etc) that we can't feasibly obtain via targeted testing.

We should be sure to do selective restores across replica sets, sharded clusters, and sharded clusters with sharded collections via modifying each magic restore suite.

We have the magic_restore_backup.js and magic_restore.js hooks that perform backup and restore respectively. We can add logic in these files to perform a selective backup and restore via a parameter we pass in via yaml. We'll need to add filtering to the backup copying, and add the list of collections to restore into the restore configuration.