E5-V: Universal Embeddings with Multimodal Large Language Models