BaseTLearner fit model_c for every treatment on the same data and predict same model on same data

**Describe the solution you'd like**
Train model_c once and predict model_c once

**Additional context**
During training, for each impact from t_groups, its own model_c is trained on the same data. This increases the duration of the training by the model.
Also, during forecasting, several forecasts are made on the same model_c, which increases the duration of forecasting.
