New `return` values for Posterior Predictive Checks

[storopoli]

We are not sure how you usually perform posterior predictive checks. But e.g. using rand(model) will just sample a NamedTuple (or maybe @torfjelde changed it to a Dict already since NamedTuple does not work as generally IIRC) of the variables, so it's not necessary to explicitly return a named tuple of variables in general.

Originally posted by @devmotion in #64 (comment)

[storopoli]

@torfjelde what is the best way to implement posterior predictive checks?
I will try to implement it accordingly in a PR