Detecting Misinformation in Multimedia Content through Cross-Modal Entity Consistency: A Dual Learning Approach