kfold cv function for LOGO-cv

[pboesu]

A common use-case for the more general mixed-effects variants will be sth like multi-year datasets and models with an annual random effect. In that case it would be more appropriate to do leave-one-group-out CV, e.g. leave-one-year-out, as environmental covariatees and the like will most likely have strong dependencies within year (if not being identical across year groups).

kfold.brmsfit might be a good place to start? also maybe look at NABBS paper CV strategy.