Add options to `plot.density_kind`?

[OriolAbril]

The plot.density_kind rcparam currently defaults to kde, and the value in there is taken at face value. Before, even when kind="kde" it would be switched back to histogram on discrete data.

I was thinking we could add an "auto" option in which case each plot would choose the marginal distribution visualization depending on data type and number of observations. We could then have ecdf as default for low number of samples, kde for large-ish samples and continuous data, histogram otherwise. Users could change it to the current kinds to hardcode the value or we could even try adding something like kde-auto/hist_auto... to indicate that type of visualization should be preferred but still switched when it makes no sense.

[aloctavodia]

This is something I really miss from legacy arviz. Just to be sure we are on the same page, the decision should be made per plot, not per figure.

[OriolAbril]

I was thinking per variable which might not be either of the two but I think it will always make sense to use the same visualizations for slices of the same variable even if they could end up on different plots