Check correctness of mapping subtyping for rest types

[jclark]

Following #993, I believe we need to consider N distinct labels that do not occur in the positives or negatives where N is the number of negatives, since for each negative we need to be able to freely choose the type of one label to avoid being matched by that negative.