Text this: Exploring Learned Surveillance Video Coding with Long-Term Reference and Adaptive Long–Short Modeling.