ChARM: Character-based Act-adaptive Reward Modeling for Advanced Role-Playing Language Agents | Read Paper on Bytez