Add aggregation method parameter to Imputers

[alex-hse-repository]

🚀 Feature Request

Add possibility to perform different kids of aggregation for seasonal strategy in TimeSeriesImputerTransform

Proposal

1. In TimeSeriesImputerTransform add parameter aggregation_method: AggregationMode = "AggregationMode.mean -- method to aggregate values in the window (see link)
2. Update method _fill to handle the logic with aggregation_method -- now it should aggregate values according the the passed method
   

   etna/etna/transforms/missing_values/imputation.py

   Line 203 in cfbfb01

   df.iloc[i] = np.nanmean(df.iloc[indexes])

3. Update the docs

Test cases

Add test for the seasonal strategy which checks that it works correctly with different aggregation_method

Additional context

No response