Temporal Logic Imitation: Learning Plan-Satisficing Motion Policies from Demonstrations