VEDIT: Latent Prediction Architecture For Procedural Video Representation Learning