Unexpected behaviour in `plot_dist` and `plot_forest` when models have different number of chains.

[aloctavodia]

from arviz_base import load_arviz_data

import arviz_plots as azp

azp.style.use("arviz-variat")

c = load_arviz_data("centered_eight")
n = load_arviz_data("non_centered_eight").sel(chain=slice(0, 2))

pc = azp.plot_forest(
    {"Centered": c, "Non Centered": n},
    var_names=["tau"],
    combined=True,  # works correctly if combined=False
    backend="matplotlib"
)

This also generates the wrong result.

pc = azp.plot_dist(
    {"Centered": c, "Non Centered": n},
    var_names=["tau"],
    backend="matplotlib"
)

There are also issues for plot_dist and kind="ecdf", the model with fewer chains does not saturate at 1 (I think it saturates at the ratio of the number chains in the model with fewer chains over the number of chains in the model with more chains)
When kde="hist" it fails with ValueError: Number of samples, -9223372036854775807, must be non-negative.

[OriolAbril]

my guess is the nans we introduce when merging heterogeneous data are not filtered out so the hdi ends up as nan which translates to no plotting. For the others there might be some deeper issues in the counting the number of samples and normalizing the results

[aloctavodia]

I can confirm that the ci not showing is related to nans passing skip_na to the eti/hdi function "fix" the issue.