Paper on Your other Left! Vision-Language Models Fail to Identify Relative Positions in Medical Images accepted at MICCAI 2025