We're thinking about how and where to add the matOptimize callback in larch-usher.

• the callback PR (here) does not seem to be incorporated in the larch-integrated branch of MatOptimize, or at least, there is no callback struct explicitly constructed for profitable_move_enumerators anymore. I don't think it's a huge change, but we should re-integrate that. It would be great if optimize_inner_loop took the callback as a parameter.
• as far as callback formats are concerned, this is something we might want to declare and pass to the optimize_dag_direct function, right? I am probably making this more computationally expensive (and less pretty) than it needs to be, but my first thought was that we could create a calculate_dag_score and a calculate_move_score, and perhaps a compare_scores (generic functions) to pass to optimize_dag_direct, and we could use the first to generate summary statistics on dag as it evolves, and the second and third to generate a callback instance for optimize_inner_loop to use.
• A concrete example of how we could use this callback initially: Each node in a tree/dag (actually its parent edge) determines a 'split', which is just the union of the node's child clades. An SPR move introduces new splits along the path from source node to target node. We want to accept proposed moves according to whether they result in new splits that aren't already in the DAG. As a slightly easier way to start we could consider only whether the split determined by the parent node of the SPR target is already in the DAG.