SGDA with shuffling: faster convergence for nonconvex-P{\L} minimax optimization