Feature the 'rray' package for consistent handling of arrays

[paul-buerkner]

In several places, brms has to deal with multidimensional arrays mostly consisting of posterior samples and sometimes of data. Due to R's bad default (drop = TRUE etc.) handling these arrays is painful at times and requires a lot of workarounds. The rray package (https://github.com/DavisVaughan/rray) which is soon to be on CRAN offers a powerful and consistent way to handle arrays and it could be worth using in brms.

[paul-buerkner]

Our new posterior package (stan-dev/posterior#13) will eventually do all the handling of multidimensional arrays of posterior samples (without or without rray at the backend) and I think this issue can be closed.