Overcoming non‐convexity of the robust nonlinear output feedback problem: An adaptive linear matrix inequality solution