VISAGE: Video Instance Segmentation with Appearance-Guided Enhancement