Extremum Seeking for Nonlinear Uncertain Systems: A Small-Gain Synthesis