Allow Dirichlet prior on weights for Gaussian prior in hierarchical analysis
In the hierarchical analysis you can choose a multi-model Gaussian prior, with each mode being assigned a weight. Currently you can add an individual prior for each weight, but they are independent and don't add up to 1. This MR changes that by allowing you to either use a fixed set of weight, or use a conditional Dirichlet prior for all weights (which will be one less than the number of modes). You cannot have a prior on just a subset of weights.
Closes #32 (closed).