Text this: Exploring Downscaling in High-Dimensional Lorenz Models Using the Transformer Decoder