Define generalized likelihood-function for BMU

[jgrashorn]

It would be useful for Bayesian updating to have some generalized likelihood as input to the Bayesian updating. The main motivation would be to make the definition easier for the user. This could also be a (abstract) struct s.t. we can implement different likelihoods, i.e. Gaussian or Bhattacharyya. Or formulate a struct with data and model values as input, which then returns a function that can be used in Bayesian updating.