[BUG] convert to period bug in _StatsModelsAdapter

[ciaran-g]

Describe the bug

For panel data with datetime indexes, converting datetimes to period index causes an error in the predict stage in the _StatsModelsAdapter models

To Reproduce

from sktime.datasets import load_airline
from sktime.forecasting.exp_smoothing import ExponentialSmoothing
import pandas as pd
y = load_airline()

# create dummy index with hourly timestamps and panel data by hour of day
y.index = pd.date_range(start='1960-01-01', periods=len(y.index), freq='H')
y.index.names = ["datetime"]
y.name = "passengers"
y = y.to_frame()
y['hour_of_day'] = y.index.hour
y = y.reset_index().set_index(['hour_of_day', 'datetime']).sort_index()

forecaster = ExponentialSmoothing(
    trend='add', sp=1
)  
forecaster.fit(y)
forecaster.predict(fh=[1])

Expected behavior

For each panel predict one step ahead

Additional context

I will address this issue 👍

Versions

Python dependencies:
pip: 23.1.2
sktime: 0.24.1
sklearn: 1.2.2
skbase: 0.6.1
numpy: 1.24.3
scipy: 1.10.1
pandas: 2.0.2
matplotlib: None
joblib: 1.2.0
numba: None
statsmodels: 0.14.0
pmdarima: 2.0.3
statsforecast: None
tsfresh: None
tslearn: None
torch: None
tensorflow: None
tensorflow_probability: None

[fkiraly]

We should add a test for all forecasters - that could be done in utils._testing.scenarios_forecasting, in forecasting_scenarios_extended