Text this: Stochastic Variance Reduced Primal–Dual Hybrid Gradient Methods for Saddle-Point Problems