Reading-strategy Inspired Visual Representation Learning for Text-to-Video Retrieval