refactor NonTargetWeed-Labeler

[mkutu]

The idea is to use this same code base to train other version of a classifier to develop somewhat of a model zoo that can be pulled for this specific task and for specific species or locations.

• move code from NonTargetWeed-Labeler to a new repo in psa called SemiF-NonTargetWeedClassifier
• implement hydra
• refactor image_labeler.py to that it's either labeling new images or labeling images that have been predicted by a "teacher" classifier
• refactor predict.py so that it's the model is inferencing on multiple images instead of one by one
• add confidence values to csv in predict.py in a column called "PredictedTargetWeed_Conf"
• refactor move_files.py