Transferring CLIP's Knowledge into Zero-Shot Point Cloud Semantic Segmentation