Asymptotically Fair Participation in Machine Learning Models: an Optimal Control Perspective