Unifying Attention and Diffusion with Kan Extension Transformers: Structured Deep Learning with Diagrammatic Backpropagation