Higher-order Lie bracket approximation and averaging of control-affine systems with application to extremum seeking