Learning Lifted Action Models From Traces of Incomplete Actions and States