On Combining Expert Demonstrations in Imitation Learning via Optimal Transport