[BUG] The ndarray wrapper in cudf.pandas doesn't have __array_interface__.

[trivialfis]

Describe the bug
The ndarray wrapper in cudf.pandas doesn't have __array_interface__, but has __cuda_array_interface__ instead. I assume this is caused by it being backed by cupy. Discovered in #15906 (comment) .

Steps/Code to reproduce bug

# file: reproduce.py
import cudf

print(cudf.__version__)         # 24.08.00a101

import pandas as pd
df = pd.DataFrame({"a": [1, 2, 3], "b": [4, 5, 6]})
arr = df.values

print(arr.__cuda_array_interface__) # good
print(arr.__array_interface__)  # AttributeError

Run:

python -m cudf.pandas ./reproduce.py 

Expected behavior
I don't know what to expect in this case. This might actually help XGBoost to work with cuDF pandas wrapper since we can directly fetch the CUDA memory, albeit in a hacky way exploiting an unexpected difference between cudf.pandas and pandas. It's still a bug instead of a defined behavior. I will leave it to cuDF devs to decide what to do with it.

Environment overview (please complete the following information)

• Environment location: Bare-meta
• Method of cuDF install: conda

Environment details

• cuDF: 24.08.00a101