Replace RC examples with bbox-annotated first frames

Initial MedVidQA demo: 8 medical video tasks with Qwen3.5-VL predictions