Expose EvaluationResponse from LLM

Expected Behavior

The Fact Checking Evaluator and the Relevancy Evaluator should also give the reasoning back. This is required for the testers to understand why the test is a pass. This gives transparency and also supports audits to be recorded as to why the score was pass one day and failed the other.

Current Behavior

The current evaluation framework also gives me a score and a success/fail. While this serves the purpose but creates confusion as to why its a pass or a fail. I tried passing my own prompt but more times the change of prompt made the yes.ignorecase() to fail and to pass me a fail, though the LLM was passing the reasoning as a pass.

Context

Because of the current behaviour I could not use the evaluators out of the box. As a workaround for now I created custom evaluators. However I would like to contribute to enrich the current code to satisfy these kind of extra cases.

Here is how I want to do it

1. There is already a feedback that is now hardcoded to an empty string. I would like to use that field to pass back the reasoning by giving a way for the users to pass their custom prompt to wrap the reasoning as feedback
2. Also my idea would be to expose the whole LLM response in the evaluation response so that users can write their custom prompts and can use the response in the way they want if they do not want to use the out of box score and success/fail.

Start in the evaluation framework at the Fact Checking and Relevancy Evaluator entry points, then trace how EvaluationResponse is assembled and where feedback is hardcoded to an empty string. Determine how custom prompts and the complete LLM response should be represented; done means evaluators can return reasoning and expose the LLM response without losing the existing score and success/fail results.