move score coefficient and optimal weight tree counting

[willdumm]

This PR contains two improvements to larch-usher:

• adds options --move-coeff-nodes and --move-coeff-pscore which are integer coefficients with which to calculate the score of a move in the callback. A move's score is computed here: https://github.com/matsengrp/larch/blob/dcc0a259c21a2884e60293e4c4ee32462584dc63/tools/larch-usher.cpp#L149 where move_score_coeffs is a pair containing the new node coefficient first, and the parsimony score coefficient second.
• Adds SubtreeWeight method MinWeightCount which allows counting of minimum-weight trees without trimming, or using the much less efficient WeightAccumulator. A better implementation may be possible when trimming which returns a view is implemented. Alternatively, it would be great to be able to compute an arbitrary WeightOps hypothetically on the trimmed DAG (this is essentially an implementation of this idea where the arbitrary WeightOps is TreeCount.)

[willdumm]

In the last commit I added MinWeightSampleTree to sample a minimum-weight tree, given any SubtreeWeight... This could be extended to uniformly sample a min weight tree using cached_subtree_counts_, which might be more useful. I also added an option to use this min-weight sampling, and I added elapsed seconds to the logger.

I haven't had time today to write a test for minimum weight sampling.