When using an Auto model, why does it clobber my early_stop_patience_steps? https://github.com/Nixtla/neuralforecast/blob/main/neuralforecast/common/_base_auto.py#L140-L143