Ensure correct probability masking if evaluating with non-present class

Summary

This MR introduces the following changes

• Fixing an issue in the validation plots that taggers who don't have the output for a certain class (taus for example) are skipped and not plotted.
• Ensure correct masking with zeros for classes for which the tagger (which is to be evaluated) is not trained. The tagger output probability values for this class are masked with zeros so we can evaluate (for example) the tau rejection on DL1dv01 (which is not trained on taus).

Conformity

• Changelog entry
• Documentation
• Development guidelines
• Style guides