Fusion-Eval: Integrating Evaluators with LLMs