Diagnostics: partitioning filtering and naming

[jerrymhuang]

Give users the ability to rename their variables through separating the filtering and naming processes in dict_to_arrays and preprocess.

Related to #257 .

[paul-buerkner]

thank you! Could you please provide a small code example for how users would use this new feature?

[jerrymhuang]

Update: the partitioning between filtering and naming is now even clearer!

Here are two examples of the updated usage for diagnostics (as already indicated in the Linear Regression notebook):

Example 1: plot_samples_2d

# Let's say we only want to plot the beta prior samples...
filter_keys = ['beta']
variable_names = [r"$\beta_0$", r"$\beta_1$"]

f = bf.diagnostics.plot_samples_2d(
    samples=prior_samples, 
    filter_keys=filter_keys, 
    variable_names=variable_names,
    context="Prior"
)

Example 2: plot_recovery

# Let's say later we want to plot recovery for all parameters...
filter_keys = ['beta', 'sigma']
variable_names = [r"$\beta_0$", r"$\beta_1$", r"$\sigma$"]

f = bf.diagnostics.plot_recovery(
    post_samples=pdraws,
    prior_samples=val_sims,
    filter_keys=filter_keys,
    variable_names=variable_names
)

The above examples are streamlined to all diagnostics utilities that require preprocess.