Multi-Object Tracking Integration for OpenVINO Training Extensions (OTX)

OTX supports 15+ detection models but has no built-in tracking.Every real-world video application (conveyor belt counting, asset monitoring, dwell-time measurement, traffic analysis, sports analytics, etc) needs stable object tracking across frames. This project integrates Multi-Object Tracking into OTX, adding five tracker implementations (ByteTrack, SORT, DeepSORT, OC-SORT, BoT-SORT), a SAM-2-inspired memory bank for re-identification, MOT17 benchmarking, and CLI support along with documentation. The design is decoupled: a new OTXTracker base class owns all temporal state while existing OTX detection models remain untouched.