We rely on a collection's sharding key having enough distinct values so to split keys across chunks and on autosplitting working the large majority of times.

If, for some reason, a very large chunk forms, we want to detect it.

One point we can do that is when we get all the disklocs during a migrate. If the total size or # of objects is too big we want to fail the migrate and force a split.

If we can't split them, find a way to prevent attempting to move it repeatedly.