Propose attachment uniformly over tree length, rather than uniformly over edges about sts

[cmccoy]

This is implemented here for unguided proposals.

Guided proposals are a bit trickier. So far I've tried discretizing the tree in a very simple way - the total branch length gets stretched into a line, likelihood is evaluated at N evenly spaced points along the line.

Since there may be multiple points proposed on a single edge, I believe the forward proposal probability should be the sum (in non-log space) of likelihoods for each edge (implemented here, used in the proposal).