Reassess usage of dclab.dfn.scalar_feature_names / make sure the user can see ml_score features when exporting statistics

[paulmueller]

...because ideally, we would also cover machine-learning features that are not made available via MachineLearningFeature. E.g. machine-learning features may be inherent in .rtdc datasets.

[paulmueller]

I took a look and refactored some of it in 5513266

The only place where ml_score features are not included is comp_stats.py in the case of computing statistics for data in external directories. Here, the entire directory tree would have to be searched for the machine learning features. We could include a button for this to include them.