Diffusing Surrogate Dreams of Video Scenes to Predict Video Memorability