If there are multiple threads inserting for the same donor shard, we can end up with writing an older batch later than writing a newer batch and break the resume token.

Check BF-29653 for more context