create 'reverse non-centered' phylogenetic models in stan

[rmcminds]

naively centered parameterization of a multilevel phylogenetic model leaves every node centered, e.g. directly representing evolutionary process. naively noncentered adds an effect to each descendant node based on its parent. in theory, it is likely that the best parameterization would be one where the tips are centered but internal nodes are not. e.g. use raw parameters for tip values, because data is likely to inform them strongly, and use predictions of ancestral states as an offset for higher nodes, so that their raw parameters refer to deviations from the prediction. I think you could use the same code as a centered parameterization for augmenting lp__, but would need to construct transformed parameters as the prediction + deviation to use in that code.