R2D2 does not account for scale of the data

[n-kall]

Currently, it seems that the R2D2 prior does not take into account the scale of the covariates. In other words, it seems to assume that the predictors (and outcome?) are standardized. However, I haven't found anywhere in the documentation that this is stated.

I think it could be helpful if this is added to the the documentation on R2D2, and perhaps it should be an option to scale the lambdas as in the R2D2M2 paper (equation 7).

[paul-buerkner]

Good point. Same goes for horseshoe actually, I think(?) An autoscaling option for the lambdas would indeed be nice. I assume the same thing could be easily done for horseshoe too. @avehtari what do you think?

[avehtari]

Scaling (that can be turned off) would be useful

[paul-buerkner]

I think we already have the autoscale option that currently scales (only) by sigma. I think making autoscale also scale according to the predictors' scales would be great.