Closes #5096: unit tests for inherited functions in arkouda extension array

[ajpotts]

PR: Comprehensive Pandas ExtensionArray Inheritance Tests + Categorical Stubs

Summary

This PR introduces a comprehensive test suite verifying that Arkouda’s
ArkoudaArray, ArkoudaStringArray, and ArkoudaCategoricalArray correctly inherit behavior from pandas.api.extensions.ExtensionArray.

Tests ensure:

1. Arkouda Does Not Override pandas EA Methods

For each method, the suite asserts:

• the method is not defined in Arkouda
• MRO resolves to pandas’ ExtensionArray
• the inherited behavior matches pandas where supported

Verified inherited methods include:

_fill_mask_inplace, _formatter, _from_scalars, _get_repr_footer,
_hash_pandas_object, _putmask, _rank, _repr_2d, _values_for_argsort,
delete, dropna, insert, isin, map, ravel, repeat,
searchsorted, shift, tolist, transpose, unique, and more.

2. Behavior-Level Parity With pandas

Tests check:

• value equality
• dtype agreement
• NaN/NA handling
• scalar vs list-like inputs
• matching behavior for numeric, string, and categorical EAs

3. Known Unsupported Behavior Is Explicitly Tested

Some pandas behaviors require dtype conversions Arkouda does not support yet.

Locked-in expected behaviors:

• ArkoudaCategoricalArray.map → xfail
• ArkoudaCategoricalArray.shift → raises NotImplementedError
• _repr_2d on numeric EAs → raises TypeError

4. Added NotImplementedError Stubs for Categorical Internals

Stubs added for pandas Categorical-internal helper APIs:
_set_categories, _replace, _reverse_indexer, _values_for_rank,
add_categories, rename_categories, remove_unused_categories, etc.

5. Test Coverage Improvements

Covers:

• unique() w/ NaNs
• hashing parity
• ranks w/ ties
• deletion, insertion, ravel, repeat
• searchsorted for strings & categoricals
• tolist() alignment with pandas EA semantics
• map() for strings, xfail for categoricals

Motivation

Previously:

• Many inherited EA behaviors were untested
• Missing documentation of unsupported categorical operations
• Risk of silent regressions from upstream pandas changes

This PR establishes a correctness baseline.

Examples

import arkouda as ak
from arkouda.pandas.extension import ArkoudaArray
import pandas as pd

arr = ArkoudaArray(ak.array([10,20,10,30]))
pd.util.hash_pandas_object(pd.Series(arr))

from arkouda.pandas.extension import ArkoudaStringArray
arr = ArkoudaStringArray(ak.array(["a","b","a"]))
arr.unique().tolist()
# ['a', 'b']

from arkouda.pandas.extension import ArkoudaCategoricalArray
arr = ArkoudaCategoricalArray(ak.Categorical(ak.array(["x","y","x"])))
arr.tolist()
# ['x','y','x']

Test Commands

pytest tests/pandas/extension/ -q
pytest -k "Map" -q
pytest -k "Categorical" -q
pytest -k "SearchSorted" -q

Closes #5096: unit tests for inherited functions in arkouda extension array

[1RyanK]

Looks good!