Dense Dynamics-Aware Reward Synthesis: Integrating Prior Experience with Demonstrations