This PR further improves the diagnostics module:

• Introduces a basic VariableArray class to carry around variable_keys and variable_names with an array containing posterior draws. This class is only meant for internal use for now and we can see if extending it in the future could be sensible.

• Further improves posterior pairs plots to enable plotting the prior again.

For example, using the linear regression notebook, we can now do something like:

bf.diagnostics.plots.pairs_posterior(
    estimates=post_draws, 
    targets=val_sims,
    priors=sim_draws,
    dataset_id=0,
    variable_names=par_names,
)
plt.show()

to show both posterior, prior, and the true parameter values. The previous implementation didn't quite work in terms of correctly showing the prior.

There is one issue which I couldn't quite resolve and I would like to get your input @stefanradev93 and @jerrymhuang: Currently, the posterior is always plotted behind the prior on the diagonal. I tried to change that via hue_order for example but it either breaks the off diagonal plots or mislabels prior and posterior. It kind of looks like a bug in seaborn to me as I cannot imagine this to be intended. Any ideas would be highly appreciated.