Draft: Generalised Dirichlet prior for arbitrary ranges

These changes should allow users to define Dirichlet prior elements on a support different from 0 to 1. The Beta prior within bilby already has this capability, so the changes here should be sufficient, I think. This is handy because otherwise users would have to reparametrise their problems to match this defintional requirement. I will also aim to add some tests.

@colm.talbot since you implemented this originally, do you agree that this would work?