Multiplier convergence in trust-region methods with application to convergence of decomposition methods for MPECs