--n is not always respected

[BobBorges]

Often the program samples a file with no diff, this generates a user warning, but then we end up with --n - 1. Perhaps instead of/in addition to the user warning, we need to select another file until the sample size == the --n input.

[ninpnin]

I think this is the case when you sample more than 1 edit per file, and then there is only one diff to be sampled.