Automatically check input of anomaly detectors

[LouisCarpentier42]

Currently, the BaseDetector does nothing with regards to .fit(X, y) and .decision_function(X). When implementing a new detector, the user has to do all the checks manually (is valid array, transform to np array, check if fitted). The beginning of each implemented .fit(X, y) and .decision_function(X) is therefore almost identical.

This issue can be resolved by taking an approach similar to the Preprocessor or Metric classes, with additional methods ._fit(X, y) and _decision_function(X), which effectively implement the anomaly detector. The original .fit(X, y) and .decision_function(X) then simply perform the necessary checks and formatting, and then call these new functions.

In addition, a method is_fitted() could be added to the BaseDetector, which returns whether the detector has been initilialized. This can be done by a simple boolean property or by checking if all properties with a trailing underscore have been instantiated.