Support Dirichlet prior distribution

Adding dirichlet to the prior shape option for #568 (closed). Needs to be supported in estimation as well.

@MichelJuillard, I think you also need to support this for the ms-sbvar C back-end. See writeCShape in ComputingTasks.cc

@houtanb As far as I understand it, the Dirichlet prior is a multivariate prior. Up to this point, we only have independent univariate distributions. Hence, there is no syntax in place for specifying a joint prior. Did you already have some syntax in mind?

@JohannesPfeifer you're right for the general case. @stepan-a created a syntax for this (See #824). If I remember correctly, and @MichelJuillard can correct me, the transition probabilities in DMM have Dirichlet distributions, so this option is actually just redundant.....

removed milestone

marked this issue as related to #226

marked this issue as related to #568 (closed)

8956bdde was the first commit.