Active learning workflow

[JorisRoels]

Include some kind of flag that enables active learning mode. This should only be available if (a small set of) annotations are already available. This will do the following:

• Infer the complete volume with the best performing model
• Select a number of unlabeled samples from the volume based on some criterion (maximum entropy should suffice for starters)
• The user should annotate the samples
• The model is retrained with the extended annotations