On the existence of optimal partially observed controls