Video IO, annotation, and demos

• process_video / process_video_list / process_folder (core/io/process_video.py) run a tracker over video files and save per-video (frames, points, 3) arrays as .npy (or JSON) via DataStore. The list/folder forms open all videos simultaneously and call process_batch once per frame, a single GPU call per model per frame, suiting synchronized multi-camera recordings; they recommend creating the OnnxSession with batch_size equal to the video count.
• KeypointAnnotator (core/annotation/keypoint_annotator.py) draws any Observation given per-stage visual schemas (connections, colors, optional colored connection groups, box overlays that distinguish "detector ran" from "bbox reused"). A dedicated ChArUco annotator also exists.
• The skellytracker console script (python -m skellytracker) opens a live webcam demo: --tracker {mediapipe,rtmpose,aruco,charuco}, --camera, --rotate, --list; each detector's run_demo module exposes finer options.