+ Skeleton type is a little more fleshed out
+ Lots of logic taken out of the mediapipe file and moved into the types
themselves for re-use
+ Re-aranged mediapipe transforms to their own dir
+ headlook-from-3d transform now outputs a skeleton + calculates roll
correctly
No idea why only sometimes this is an issue - maybe its a sign that
wrong sample rate is in use somewhere and I should handle it
differently? Idk, not a priority rn
This impliments most of the abilities available previously,
missing most notible pupil detection atm.
This also currently uses a completely different tracking algo,
mediapipe's landmarks module, but by the nature of this system
is in no way limited to that!
Derek