Text this: Dual-Distillation Vision-Language Model for Multimodal Emotion Recognition in Conversation with Quantized Edge Deployment.