`C()` operation in formulas should inherit `cat_missing_method`

[MatthiasSchmidtblaicherQC]

When using C in a formula to explicitly convert to a categorical, cat_missing_method is ignored and set to default "fail" instead:

# %%
import pandas as pd
import tabmat as tm
# %% '4.1.5'
tm.__version__
# %% whether "x" is already a categorical does not matter
df = pd.DataFrame({"x": ["a", "b", pd.NA]}) # pd.DataFrame({"x": pd.Categorical(["a", "b", pd.NA])})
# %% result as expected
tm.from_formula(formula="x", data=df, cat_missing_method="zero")
# %% result as expected
tm.from_formula(formula="x", data=df, cat_missing_method="convert")
# %% # raises ValueError: Categorical data can't have missing values if cat_missing_method='fail'.
tm.from_formula(formula="C(x)", data=df, cat_missing_method="zero")
# %% # raises ValueError: Categorical data can't have missing values if cat_missing_method='fail'.
tm.from_formula(formula="C(x)", data=df, cat_missing_method="convert")

[MatthiasSchmidtblaicherQC]

This turns out to be an intended pattern: the global cat_missing_method is not recognized by C so the user can specify different behaviours for different categoricals:

# %%
import pandas as pd
import tabmat as tm
# %%
df = pd.DataFrame({"x": ["a", "b", pd.NA], "y": ["u", "i", pd.NA]})
# %%
tm.from_formula(formula="C(x, missing_method='zero') + C(x, missing_method='convert')", data=df)

See also the tutorial.

[stanmart]

This turns out to be an intended pattern

Hm, this might be an unintuitive choice nevertheless. Would it be better if C() inherited cat_missing_method's value by default? (if it can be implemented in a clean way)

[MatthiasSchmidtblaicherQC]

Yes, I think that inheriting from the global cat_missing_method in the case that the argument to C() is not specified would be more intuitive.

[stanmart]

Fixed in #502, C() now inherits the cat_missing_method's value by default. The example above works as expected in tabmat 4.2.0.