[FEA] center implementation for rolling window

Hi,

I am currently doing some feature engineering on a timeseries. My index is a datetime and I want to apply a rolling window.
FYI the pandas code is running.
`df['feature'] = df['feature'].rolling(f'{120}s', center=True, min_periods=1).sum()`

which leads to the following error:

`NotImplementedError: center is not implemented for offset-based windows`

basically, I want to change to cudf for runtime optimization. As I have a datetime as an index, rolling operation with centring is much appreciated due to the fact of handling missing values, just shifting by timestamp is not trivial as discussed [here.](https://github.com/pandas-dev/pandas/issues/20012)

I would appreciate if you could take a look. (Maybe it occurs due to some changes in the API interface)