Add option for pseudovariance in R2D2 prior

[avehtari]

For Gaussian family, R2D2 scales with sigma. For non-Gaussian families, the scaling is fixed to 1. In case of non-Gaussian exponential family, RHS paper https://projecteuclid.org/euclid.ejs/1513306866 recommends scaling with pseudovariance. It would be nice to have an option to set that pseudovariance, as for example in case of Bernoulli model, the same prior on latent space R^2 leads to different priors on outcome probabilities depending on the used pseudovariance.

[paul-buerkner]

Good idea. Will check it out!

[wds15]

The same would be very useful for the horseshoe priors as well, I guess, no?