Painless Stochastic Gradient: Interpolation, Line-Search, and Convergence Rates