Discretizing the Fokker–Planck equation with second-order accuracy: a dissipation driven approach