CogVis: Must Open-Vocabulary Change Detection Perceive the Scene Anew for Every Query?