Filterset API version 0

This MR relates to #1 (closed) and gets us another step forward. We now take some description of filters from the front end and can pass it with a query to an OpenAI-style LLM, get a response, and return it.

Here's the next level of detail in what this patch set includes:

• Put sample and exemplar data in json files on disk -- this will help with mocking I think
• Convert front-end descriptions of filtersets into OpenAI function -- this probably doesn't work very well, but is a place from which we can start to iterate.
• Iterate on testing filterset endpoint -- after mocking, we can start making this more real
• Add a rudimentary mkdocs-based manual
• Minor cleanups