FMBench support for RAG + Model evaluations

This version contains [Work in Progress] code for evaluating both:

1. Majority Vote: Using a Panel of LLM Evaluators to check for RAG eval on whether a given candidate model output is correct or incorrect.
2. Average Pooling: Using a Panel of LLM Evaluators to evaluate candidate model responses using 'user-defined' subjective evaluation criteria.