Multimodal Emotion Recognition with Vision-language Prompting and Modality Dropout