Revise script/task to refine logic for updating model

[rlskoeser]

refinements for #39 and #42

[cmroughan]

Looks like we're passing the usual tests for training from scratch, refining, and refining + overwriting on segmentation and transcription jobs! Cleanup, task messaging, and model uploading (or not uploading, if a poorer model would overwrite a better one) is working correctly.

I committed a change to the overwrite handling to address the case where "overwrite" is selected, kraken finds a _best.mlmodel, but kraken's best is lower accuracy than the starting model. In that case I think it makes sense to continue following the logic of not overwriting the higher accuracy model.

Also added a commit to handle the edge case where a queued job is cancelled in MyDella, since this results in no .out file being created and was causing the script to error out on a FileNotFoundError.

I'm setting up one last round of tests to run tonight and will double check the results in the morning but I think we're looking pretty good here!

[cmroughan]

Results of last night's tests successfully showed the behavior we want to see. The various data is also being sent to Task messages correctly and this continues to be very helpful for testing and troubleshooting. I also ran tests to confirm that when one person shares a model with another person, the second person's user permissions still do not permit overwriting the original person's model (though they can use it for non-overwrite refining as usual).