scale covariance matrices based on phylogeny so determinant = 1, not to correlation matrices

[rmcminds]

exp(mean(log(eigen(cov_raw)$values)))

determinant is 'generalized variance' and will take shape of tree into account, e.g. if there are two extremely divergent lineages, each with large but extremely recent diversifications (pom-pom tree), then a single phylogenetic effect using correlation matrix as prior would essentially be saying we expect the two lineages to be different, but no difference within the lineage. However, it might be better to have the effect allow for variation among all the tiny tips of the lineages, and expect the variance of the effect separating the lineages to be much larger than a normal effect?

phyr does this. think about how it interacts with e.g. adding OU effects.