[Base_recurrent] Scale `base_recurrent` models with rolling window statistics

[cchallu]

Description

Currently, base_recurrent models (RNN, LSTM, GRU, TCN, etc.) scale time series using the global statistics for each time series (on the train set). This might reduce the performance of models on non-stationary time series.

We want to add the option on base_recurrent class to scale based on statistics on the recent history (similar to the scaling in the base_windows class).

Use case

No response