Scalable First-Order Bayesian Optimization via Structured Automatic Differentiation