WIP: Feature: add a qna evaluation for post training checkpoint evals

[nerdalert]

• Give the user the ability to evaluate their QnA from the dashboard against a checkpoint on the server. This requires the same API server being used in the end to end fine-tuning demo.
• Add endpoints for a checkpoint listing and the execution of a qna eval.
• The backend go api server will call a container that loads the requested checkpoint and qna.yaml into vLLM and queries the checkpoint with the seed questions from the QnA to get a sense of the fidelity of the checkpoint from the orignal seed examples.

Tasks:

• Remove hardcodes for qna and checkpoint paths via ENVs.
• Add the option to side by side the qna and the eval results for easy comps.

Quick snippet to give a general idea. Will change with the above tasks.