General refactor to align with curation

As a follow-up to the curation refactor work, it would be good to investigate and refactor the view/service separation in training and prediction (and the rest of the application). Known areas for refactoring:

1. currently, training and prediction's view classes are responsible for starting threads, which doesn't align with our design goal of having services be responsible for any long running / async tasks
2. training and prediction still use the custom publisher/subscriber framework instead of pyqt signals
3. training and prediction are not using the task executor abstraction for async work

This is a rather large ticket, and should be treated as low-ish priority, but it _should_ be done at some point in stages because the longer we wait on it, the bigger the change will have to be.