SPANER: Shared Prompt Aligner for Multimodal Semantic Representation