Topological Alignment of Shared Vision-Language Embedding Space