Use GPLikelihoods.jl

[theogf]

• Use the GPLikelihoods.jl package
• Reuse the existing likelihoods and links. And take advantage of AbstractLikelihood
• Do some type piracy but allow to have multiple models given the same likelihood. For example BernoulliLikelihood with LogisticLink or SVMLink. The constructors stay the same
• Add an example specific for the HeteroscedasticLikelihood
• Generally use Tuple instead of Vector