While discussing T330686, we figured that our delta algorithm assumes that the previous run was successful.
In this task we want to change this so that the delta is made againt whichever is the latest snapshot available in Hive. This will make the job robust against intermittent failures.