Get per-prompt scores from DEMO set

[mikek-mlcommons]

The DEMO prompt set is useful for tuning if we can provide per-prompt/response scores.

[mikek-mlcommons]

• Audiences: MLOps engineers, Data Scientists
• User story: "As an MLOps engineer responsible for model tuning and evaluation, I want AILUminate to see Demo prompt scores per-prompt/response so I can understand which kinds of prompts score what kind of violations and tune the models I evaluate."
• Explain how to use the Demo set with the official evaluator to understand something about how the benchmark grades responses, so they can tune their models.
• Create a tool that makes it easier to get/read/interpret per-prompt responses from the Demo prompt set
• Design a report that describes the high false-violating rate, why it behaves that way, and what the tradeoffs led to its design that way (tl;dr: it's better to mark something as violating, and to review the violation in other ways, than to miss an actual violation at the benchmarking stage).