Real-Time 3D Vision-Language Embedding Mapping