Storia: Beyond Next-Token Prediction: A Performance Characterization of Diffusion versus Autoregressive Language Models — Warptech Lab News