Describe the bug

# t.py
df = pd.DataFrame({"a": [None, 1, 2, None, 4, 6, 11]})
print(df)
print(df['a'].rolling(2, min_periods=1).sum())

Then run python -m cudf.pandas t.py

Expected behavior

0     NaN
1     1.0
2     3.0
3     2.0
4     4.0
5    10.0
6    17.0
Name: a, dtype: float64

Actual behaviour:

0     NaN
1     NaN
2     3.0
3     NaN
4     NaN
5    10.0
6    17.0
Name: a, dtype: float64

spotted in Narwhals

Environment overview (please complete the following information)

• Environment location: [Bare-metal, Docker, Cloud(specify cloud provider)]
• Method of cuDF install: [conda, Docker, or from source]
  • If method of install is [Docker], provide docker pull & docker run commands used

Environment details
Please run and paste the output of the cudf/print_env.sh script here, to gather any other relevant environment details

cudf '25.02.00a289'

Additional context
Add any other context about the problem here.