On optimal control of forward–backward stochastic differential equations