Convergence Guarantees for Stochastic Subgradient Methods in Nonsmooth Nonconvex Optimization