Comprehending Spatio-temporal Data via Cinematic Storytelling using Large Language Models