Adding possibility to evaluate classes the freshly trained tagger is not trained on

Summary

This MR introduces the following changes

• Adding the new evaluation option extra_classes_to_evaluate. Here you can add classes for which the tagger is not trained. When the evaluation is run, these extra classes are also loaded and the tagger is evaluated with them. With this, you can test the behaviour of the freshly trained tagger for jets it is not trained for.
• Adding a fix for the confusion matrix. If a flavour is not used for the matrix, although the network was evaluated on it, the plotting would return an error. This is now fixed by removing all unwanted flavour jets from the matrix plotting.
• Making the rejection calculation more robust. When a ZeroDivisionError is given, the value added to the rejection dict is now NaN.

Relates to the following issues

• Closes #129 (closed)

Conformity

• Changelog entry
• Documentation
• Development guidelines
• Style guides