AI Tired of Tuning Learning Rates? Meet DoG: A Simple Parameter-Free Optimizer Backed by Solid Theoretical Guarantees