order NNIs for walk by parsimony scores

[chrisjenningsshaffer]

In the current version of the proof of concept NNI search, we rank trees by their likelihood, but this can be replaced with something else. Let's try using the parsimony scores as given by matsengrp/gctree#93.

The results should be compared against the likelihood ranking. We'd like to see parsimony scores eliminate the need for some of the likelihood evaluations.

For another time, we could instead try using a distance based criterion (https://gite.lirmm.fr/atgc/FastME/).