Idea: add statistical labeling to elm-test

Don’t think that would be very valuable, since whether or not one gets the warning is pure luck as is whether the warning is a false positive.