Graphical Convergence of Subgradients in Nonconvex Optimization and Learning