On Quadratic Convergence of DC Proximal Newton Algorithm in Nonconvex Sparse Learning