[question](phase1): should the paco mask of the image be combined with the bbox layer for music to strip line ink and noise to aid the downstream IC?

Plonking this here for thoughts and also so I don't forget. I'm wondering if we can pull some hijinks with the remaining mask that could be produced by pacos and combined with the bboxes for music (essentially so we know where we expect and don't expect ink) to help remove noise from the page downstream. This could additionally be deployed in a secondary reclassification by YOLO, or if we implement some kind of BGR as part of the phase 1 flow.

As we encountered with Rodan, adding this kind of component can have a variable impact on performance, particularly given the breadth of manuscripts we're hoping will be able to be passed successfully through the omr process, as well as just increase the complexity of the project as another variable to create, insert, and maintain.