This is needed for proper sharding support.
Right now it's hardcoded to use an ObjectId which means it's ever growing and hits the same shard.

Also it would allow to have a unique constraint on filename, if it's used as _id.