A simulation–approximation approach to sample size planning for high-dimensional classification studies