Auto-Sklearn: add parameter for enabling anomaly detection

[johnantonn]

The anomaly detection flow for Auto-Sklearn disables the calculation of f for the training set, where we assume that the process will be unsupervised.

In order for the original test suite to pass, it would be better to add a switch-parameter for enabling anomaly detection on Auto-Sklearn. That way, the original tests should pass without any issues.

[johnantonn]

For now, a pseudo-check was added to train_evaluator.py and automl.py to enable seamless integration with the original version of Auto-Sklearn. Later, this check needs to be controlled by a flag that will indicate normal execution or unsupervised anomaly detection for training PyOD classifiers.