One-click augmentation

[cdbethune]

To get discussion on this going, I think this could look like the following:

1. User clicks on "label dataset". At this point we can do a quick threshold test to check how many labels they've generated and warn them if there are too few based on some threshold.
2. Laelled dataset is saved as a temporary dataset. Cloned dataset is deleted.
3. Classifier is automatically run on the temporary dataset using ROC_AUC as the score. We check the score on the server side, and warn the user if it is too low.
4. The original unlabelled dataset is pushed through the newly trained classifier. The temp labelled dataset is deleted.
5. We move to the predictions view to inspect the now augmented dataset, at which point the user can save it.