Capturing and Inferring Dense Full-Body Human-Scene Contact