Differing behavior of `extract_model_data()` for `"stanreg"` fits and `"brmsfit"`s

[fweber144]

See this comment in paul-buerkner/brms#1230: Do we want the behavior of predict.refmodel(), proj_linpred(), and proj_predict() to be the same for "stanreg" and "brmsfit" fits? If yes, how should that common behavior look like: The current "stanreg" or the current "brmsfit" behavior?

[fweber144]

I think the brms way (including the requirement that offsets and weights have to be specified as right-hand side formulas for columns in the datasets, not as separate vectors) would be preferable as it is more consistent, easier to understand for the user, and avoids the internal definition of additional "helper" data colums (which is error-prone, see this comment).