Integrate external motion estimation into trackers

Currently, external motion estimates (e.g., from MotionEstimator) are only used for visualization (MotionAwareTraceAnnotator) and not by SORTTracker or DeepSORTTracker for data association or state prediction . This limits robustness in complex motion scenarios (e.g., UAV, moving cameras). Current behavior SORTTracker.update(detections) relies solely on detections and an internal Kalman filter tracker.py:131-182 . DeepSORTTracker.update(detections, frame) adds appearance features but still ignores external motion . Desired behavior Allow trackers to accept an external motion transform (e.g., coord_transform) and use it in: Prediction: adjust Kalman predictions or directly warp predicted boxes. Association: add motion consistency to the cost matrix alongside IOU/appearance. Proposed solution Extend update with an optional motion_transform parameter (backward-compatible): detections = tracker.update(detections, frame=None, motion_transform=None)