Add reloo function

[jordandeklerk]

Do we want to add a reloo function for exact refitting for problematic observations with high Pareto k value?

This exists in legacy ArviZ using the SamplingWrapper class. We could include something very similar here with some minor modifications to the reloo function to account for the new PSIS-LOO-CV functions we have here that do not exist in legacy ArviZ.

[aloctavodia]

Yes, we should add it.