CALICO: Part-Focused Semantic Co-Segmentation with Large Vision-Language Models