Attention-Guided Integration of CLIP and SAM for Precise Object Masking in Robotic Manipulation