Selectively Apply Transforms to Images and Masks

[Pale-Blue-Dot-97]

Is your feature request related to a problem? Please describe.
Transforms are currently applied to either the imagery or the masks. This presents issues for segmentation tasks as geometric transforms of imagery results in mismatched ground truth masks.

Describe the solution you'd like
Transforms need to be marked as geometric in the YAML so they can be selectively applied to both imagery and masks via MinervaCompose.

Describe alternatives you've considered
Simply defining the same transforms for masks will often not work as random transforms will be applied randomly and with random parameters hence still resulting in differing outputs for images and masks. Simply applying all transforms to both images and masks will also represent issues as transforms that alter the spectral characteristics of the images will obviously create undefined outputs for the masks.