LOO and ELPD for non factorizable models

Hello, I would love to use Arviz.jl to do some LOO-PIT/LOO stuff, but my modesl are non factorizable, i.e. my likelihoods are multivariate normal distributions with a non diagonal covariance matrix. Here is a [paper](https://arxiv.org/abs/1810.10559) showing how to do this exactly for such a model. Is it possible to do that with the current Arviz.jl? It looks to me it is assuming the model is factorizable. If not, would anyone be willing to help with the implementation of the paper approach within Arviz.jl, such that we can merge it in the lirbary afterwards? Thanks!