Enable Bayesian Workflow by not automatically sampling posterior?

[drbenvincent]

After #22, we need to think if we need a bit of an API change. At the moment we just specify a model, model formula, and data. When we create an experiment object, it immediately proceeds to sampling.

But part of the point of switching to Bambi is to allow the user to specify priors. So we need to think about the Bayesian Workflow. In an ideal world, users would be able to visualise the prior predictive etc before proceeding to sampling.

This could simply mean that we create an experiment object, then sampling from the prior happens, but nothing else. The user could then call some plot functions to check prior predictive. Then finally call a .fit method. May need to think this through more, but something along these lines where the Bayesian Workflow can be respected would be great.

Needless to say, this would be a pretty big change at this point. But worth doing properly.