New "holistic" Mediapipe tracker: Body + Face + Hands

[golanlevin]

Google advertises a "holistic" landmark tracker that allows simultaneous tracking of Body, Face and Hands:
https://ai.google.dev/edge/mediapipe/solutions/vision/holistic_landmarker

The task outputs a total of 543 landmarks (33 pose landmarks, 468 face landmarks, and 21 hand landmarks per hand) in real-time.

[shiffman]

Thank you @golanlevin! @ziyuan-linn I wonder if this could help with some of the issues we came across related to #134 and #69?

[ziyuan-linn]

This looks promising! We can technically already do this in ml5 by creating an instance of FaceMesh, BodyPose, and HandPose each. However, having a holistic tracker that encapsulates everything could be more convenient. I can see this applied to avatar/VTuber projects.

I wonder if TFJS will have an API for this tracker. Will definitely look out for this!