Only include random effects in `estimate_grouplevel()`

[strengejacke]

Question is, do we want that here, too? or only return parameters that have random effects?

It would probably make sense to just return parameters with random effects, as this is the point of grouplevel() isn'it?

Originally posted by @DominiqueMakowski in #465

[bwiernik]

I think that's fine by default, can we include an option with the paramters argument to include everything including fixed?

[DominiqueMakowski]

Might be feature creep, doubt anyone would ever need that