An $hp$-Adaptive Minimum Action Method Based on a Posteriori Error Estimate