CLIP-Mono3D: End-to-End Open-Vocabulary Monocular 3D Object Detection via Semantic–Geometric Similarity | Synapse