feat: add JsonStorage and ability to chat with References

[gjreda]

fixes #102 and fixes #82

Note that you will need to upload some PDFs and rerun PDF ingestion first. That will create a references.json which is needed for chat interactions.

There is more backend work to do related to chat, but I've captured that in other issues (see comment below). I didn't want this PR to continue to increase in scope and also think it'd be good to have this piece in so that we can begin to demo it.

❯ yarn python

❯ ./src-tauri/bin/python/main-aarch64-apple-darwin/main chat --text "What can you tell me about hidden feedback loops in machine learning?" | jq
[
  {
    "index": 0,
    "text": "Hidden feedback loops in machine learning refer to situations where two systems indirectly influence each other through the world, leading to changes in behavior that may not be immediately visible. These loops may exist between completely disjoint systems and can make analyzing the effect of proposed changes extremely difficult, adding cost to even simple improvements. It is recommended to look carefully for hidden feedback loops and remove them whenever feasible."
  }
]

• Add tests for chat.py
• Add tests for ranker.py
• Add tests for storage.py
• Write response to stdout
• Add source filename or title to response (or create as new issue)
• Create issue to capture idea of "named threads"
• How does this handle "chat history?"
• How does this handle statements or non-questions?

[gjreda]

I have a preference for small PRs and it feel like this one is already getting pretty large. In order to move this PR forward, I've created three follow-up issues for myself.

1. Add source Reference metadata to Chat response #143
2. Capture chat history and create "named threads" for Chat interactions #144
3. Move References Storage Path into an environment variable or settings #145

I think there's an open question on how and whether we want to handle chat interactions that are not questions, but IMO that might be something that is informed after spending some time with this implementation.

[sehyod]

nit: the actual param name is query, not input_text

[sehyod]

Should we have a try except here, in case the filepath or the file content is incorrect?

[cguedes]

Everything looks good. My only comment is for the chat return type being an array instead of an object that contains an array.

[cguedes]

Shouldn't we return an object in the root instead of array. That would be better for evolvability.