Linear convergence in directed optimization with row-stochastic matrices