[ML] Missing influencer values which annotate all documents in a time bucket

The probability of a bucket value depends on two factors for certain types of function: 1) the function value and 2) whether or not the bucket is empty. If removing all data associated with a particular field value causes the bucket to become empty we only notice this is an influencer if the function value itself is anomalous (case 1) not the fact that the bucket is non-empty which is anomalous (case 2).