Show and Grasp: Few-shot Semantic Segmentation for Robot Grasping through Zero-shot Foundation Models