Move rel() to priors

[lindeloev]

Implementing #90, I now see that rel() becomes cumbersome to support henceforth (it already was) and that it becomes somewhat ambiguous for categorical predictors. Users seem to misunderstand it anyway.

• Remove all code for rel()
• Add support for prior = list(groupX_2 = "groupX_1+ DEFAULT", catY_3 = "catY_1 + dnorm(0, 1) T(0, )"). The returned parameter should be the distribution part so the JAGS code needs to be like catY_1_return ~ dnorm(0, 1) T(0, ); catY_1 = catY_1 + catY_3
• Add explicit informative deprecation error if a user uses rel().
• Update code documentation
• Update vignettes

An advantage of this approach is that it expands functionality: allows relativity across segments (not just the former), it allows percentage-wise relativity (divide), etc.