Future work: time-variable split_disjoint_nodes routine

[hyanwong]

The older the node, the more likely it is to require splitting if disjoint. I suspect that if there are disjoint nodes within (say) 10 generations of the samples, they should not be split. For inferences that have many very recent coalescence points (e.g. pedigrees) we should probably investigate whether splitting should have some time-dependency.