Text this: Transformer-based Fusion of Longitudinal Multimodal Radiomic Features from Chest Radiography and CT in COVID-19.