Spike/llm parse verif statements

[andrea-williams]

• loaded contents of one verification statement into my local database using pypdfium2 and jupyter notebook, connected to report_attachment id=9, report_version=18 (Bugle SFO)
• hard-coded the report_attachment_id in PdfDocumentContentsService
• can now interact fairly reliably with document via chat, although some times when the LLM server is just connecting there will be a front-end timeout. Retrying with the same prompt usually works.

Future Improvements:

• is there a way to avoid the front-end timeouts (considering the backend does eventually return a response, just not quickly enough for the frontend)?
• the model's response is often formatted for markdown, but the textbox doesn't render markdown nicely.
• button to download PDF does nothing
• create job to process all existing report_attachments to extract their contents and save to database
• remove hard-coding
• also submit report contents to AI as context so that VS can be compared to submitted data
• soooooo much code cleanup needed