Only the prediction is transformed back in train_and_predict

In def train_and_predict, we only call inverse_transform for the prediction dataset. This is an issue, e.g., during hyperparameter tuning. While it does not affect the values of the train_dataset (standardscaler on standard-scaled data does not change the data), it affects the standardscaler instance and hence, the transformation of the prediction dataset in the next hpam iteration. https://github.com/daisybio/drevalpy/blob/0cadbb6aef5b26aca254c05ab51ed782d9cee6e9/drevalpy/experiment.py#L958-L1006