Generalized node support

[willdumm]

Previously, we could compute node support only with respect to a uniform distribution on histories in the DAG. This PR allows node support to be computed with respect to the distribution expressed by downward conditional edge probabilities annotated on DAG edges, and provides tools for setting these edge probabilities to express an exponential distribution on arbitrary edge weights.

Questions remaining:

• All methods using probabilities now take an argument log_probabilities which determines whether log probabilities will be used. For some methods, this argument only determines how existing probabilities on the DAG should be interpreted. However, whether current probabilities are log-probabilities could be recorded by a flag on the DAG. Would this be better?
• Is there a way to make computations on (even log-) probabilities more numerically stable?

[willdumm]

This PR contains commits from the branch wd-preorder-dp

[marybarker]

The Parsimony-weighted example that motivated this generalized method is discussed here.