(phase1)[task]: yolo boxes are too big, they should predict/place themselves as tightly around the ink as possible

AKA, why is the predicted YOLO box so much bigger than the line it is placed around (see picture).

Initially, this should be investigated with the current box system. If this cannot be fixed AND performance is poor enough, experiment with a polygon-started approach (potentially EXTENSIVE work to set this up and will require a new YOLO model).

This will hopefully help staff line detection and downstream pitch finding.