Behavior of DiploidPopulation.add_mutation

[molpopgen]

When there are multiple branches satisfying the desired frequency criterion,
we currently sample from those branches proportionally to their length.

This procedure is likely biased.
For a desired interval [a,b), the weight placed on a branch overlapping
that interval should be length*(min(b,right)-max(a,left)).

(This procedure is implemented in cpp/functions/add_mutation.cc.)

[molpopgen]

So, #1026 already picked a branch from a tree according to length*(tree span).

But that is not quite right. If the desired range is smaller than the tree span, or only partially overlaps it, then we are off by not accounting for the overlap.