refactor: extract llm-as-judge logic to reusable module

add script to evaluate rag and evaluation results