Generate a quality control report when exporting the classifier

[MaksHess]

Following up on a recent discussion with @jluethi, it would be nice to provide the user with a report assessing classifier performance and potential problems with the annotated data. Here's a collection of things that could be useful:

Scoring

• Per-class F1 score (for for than 2 classes, could even be added to the status bar).
• Confusion matrix.

Annotations

• Feedback regarding class imbalance.
• Prediction uncertainty (maybe interesting to show interactively to guide the user to annotate objects at the decision boundary).
• 2D embedding of annotated points colored by their predicted class, with the annotations superimposed. This could give quick visual feedback whether only "clear-cases" were annotated.