Test of Time: Dual Averaging Method for Regularized Stochastic Learning and Online Optimization