#76 gaussian processes

[AlexRodis]

closes #79

This PR implements Gaussian Process models in bayesian-models. GPs are a very versatile and heterogeneous superfamily of models. These variant should usable in this implementation:

• Standard GP models - with response functions and arbitrary likelihoods (classification/regression)
• Deep Gaussian Processes := Chained GP models with a layer like structure, similar to Deep Neural nets
• GPs with separable kernels (including the ICM model)

Should also implement some common graphics (i.e. the mean/variance bead graph). docs-wise need to implement api docs, multiple tutorials and how-to guides and an extensive discussions section detailing these models