Fluctuation‐driven learning rule for continuous‐time recurrent neural networks and its application to dynamical system control