allow provider to report updated prompt?

[pelikhan]

We have a framework that generates prompts on the fly (https://github.com/microsoft/genaiscript) which means that the prompt files built on the fly and sent to the LLM (This works great with the custom javascript provider model and we were able to integrate nicely with promptfoo)

This is problematic for the prompt source because it always ends up being the javascript source instead of the generated prompt, which disables a number of assertions that rely on the input.

Would there be a way to also provide the opportunity for the api provider to return the prompt it actually as part of the response? With a couple pointers, I would be happy to produce a PR.

[typpo]

This request definitely makes sense - I'm thinking about the best way to implement and will provide some pointers.