pandas_dtype_strategy supports Category dtype

[cosmicBboy]

Is your feature request related to a problem? Please describe.

Currently, the pandas_dtype_strategies function in #314 doesn't handle categorical data types. To be feature-complete, we'd want to support this, with the caveat that pandera doesn't currently support PandasDtype enums with additional metadata, such as the CategoryDtype with categories and ordered information.

Describe the solution you'd like

When constructing a field_element_strategy, programmatically fetch any Check.isin checks and get the allowed_values fields from those checks. Try to infer the pandas datatype of the underlying categorical value:

categories = []
category_pandas_dtype = None
if pandas_dtype.is_category:
   for check in checks:
       if check is Check.isin:
           # get categories and infer pandas dtype
           # and remove isin checks from the checks list
    elements = pandas_dtype_strategy(category_pandas_dtype, categories=categories)

...

And then in pandas_dtype_strategy:

if pandas_dtype.is_category:
    return isin_strategy(
            pandas_dtype.String,
            strategy,
            allowed_values=kwargs.get("categories")
    )

Since series/index/dataframe strategies cast the generated data to the correct data type, this workaround should work for now.

[cosmicBboy]

This should be done after #369

[zevisert]

#369 is done, and with Annotated types, we can embed the categories into the type!

import pandera as pa
import pandera.typing as P
from typing import Annotated

class MySchema(pa.SchemaModel):
    some_category: P.Series[Annotated[P.Category, ['foo', 'bar', 'baz', 'quux'], False]]

It seems like pandera could generate strategies using these annotations, and error as it does now if there's no categories annotation. It does seem like we'd have to contribute to hypothesis first perhaps, since their pandas series strategy implementation seems to require being able to create numpy arrays with specific dtype, rather than pandas extension arrays.