Bug: final_population_cleanup mishandles mutation counts if the only preserved generation is the final generation.

[molpopgen]

Work on #828 has discovered a bug when ancient samples are used along with stopping conditions. Given that stopping conditions have been an undocumented feature and that ancient samples are pretty well-tested, this bug hasn't hurt anyone. However, it is keeping #828 from passing certain tests.

There is a back-end bug in the final cleanup code. This case was missed in tests and in practice we've developed the habit of not recording in that generation, thus never triggering the bug.

This is the first example that we used to trigger the bug:

import numpy as np

import fwdpy11


def preserve(pop, sampler):
    if pop.generation == 10:
        sampler.assign(np.arange(pop.N, dtype=np.uint32))


def stopper(pop, _):
    if pop.generation == 10:
        return True
    return False


pdict = {
    "sregions": [fwdpy11.ExpS(0, 1, 1, -0.05, 1)],
    "gvalue": fwdpy11.Multiplicative(2.0),
    "rates": (0, 0.1, 0),
    "simlen": 20,
}
params = fwdpy11.ModelParams(**pdict)
pop = fwdpy11.DiploidPopulation(100, 1.0)
rng = fwdpy11.GSLrng(90210)

fwdpy11.evolvets(rng, pop, params, 100, recorder=preserve, stopping_criterion=stopper)

However, any recorder in a simulation w/a reasonably high mutation rate that preserves the last generation should also trigger it.