Convergence of a stochastic subgradient method with averaging for nonsmooth nonconvex constrained optimization