More than ~2 billion segregating mutations doesn't give a good error

[bhaller]

MutationIndex is int32_t (any reason it can't be uint32_t to raise the roof by 2x?), and so there is a limit of ~2 billion segregating mutations at any one time. That's probably OK for now. The bug here is that when that limit is exceeded, you apparently don't get a good error message, you get a cryptic internal error. See https://groups.google.com/g/slim-discuss/c/HJygUwCAEfU/m/WqpzggOBAAAJ.