Adding generics of general Bayesian methods

We discussed at some point, that we want to have a common place where a lot of the generics should life that packages such as rstanarm, brms, etc. use. Right now, a lot of them are in rstantools but I don't think this is a good place for them as the methods are not only relevant from rstan based packages. Some methods are better suited for special packages and as such should life there, for instance, in bayesplot for all plotting related generics and in loo for all cross-validation related procedures.

Please amend this initial comment to add more methods that we may want to put into posterior.

• posterior_predict compute predictions and return an arrary of draws
• posterior_linpred compute linear predictor values and return an arrary of draws
• posterior_epred compute means of the posterior predictive distribution and return an array of draws (see https://github.com/paul-buerkner/brms/issues/644)
• predictive_error computed differences between observed and predicted responses and return an array of draws
• posterior_interval basically summarise_draws with just two quantiles
• predictive_interval basically summarise_draws with just two quantiles after calling posterior_predict

Start by reviewing the proposed generics and the existing implementations in rstantools, then compare which methods belong in posterior versus bayesplot or loo. Done means reaching agreement on the common generic set and its package ownership.