Added Evaluation for RAG

[mihir86]

Description

This PR adds a new class for evaluating the performance of RAG on QnA types of datasets. I have also fixed a small bug in the retrieval where I only consider the 'text' column for encoding. I have made it more flexible for the user.

References

• Part of Compare accuracy on the natural questions dataset with and without RAG #3
• References: https://github.com/castorini/pyserini

[neubig]

Thanks a lot! I have some suggestions.

[neubig]

Thanks @mihir86 ! You've done a lot of work on this PR, and I really appreciate it.

I think that we can generalize the typing a bit more, but let me take a stab at that in a follow-up PR.

#51

[mihir86]

Thank you! Yes, surely, do let me know.