Ou refactor

[kopperud]

proposal to change the PhyloOrnsteinUhlenbeckREML class such that it better corresponds with the FitzJohn (2012; Methods Ecol Evol) supplement

notable changes

• Change the name/suffix from REML to Pruning. Reason being, I think it's not the REML likelihood, and so calling it REML is misleading at best. The likelihoods computed using this pruning algorithm are identical to the likelihoods computed using the generalized-least-squares (or vcv) method, as long as the root treatment is the same (i.e. root state = theta). If it really were a REML likelihood, then the likelihoods would not be the same
• Remove storing of the normalizing factors, since they were set to 1.0 and were not used anyway
• Compute the log-normalizing factor for the node, instead of the normalizing factor, saving some calls to exp(...)
• Rename "contrasts" (which they were not) to "means"
• Rename "contrast_uncertantity" to "variances"
• Add some comments to explain where the equations come from

question: what to do about the related classes, for example the state-dependent OU process (which is on a different branch)?

[hoehna]

I suggest to keep PhyloOrnsteinUhlenbeckREML as an alias for backwards compatibility.

[hoehna]

Now it looks good.