Ingest

Ingest text notes and/or files into the farm's knowledge base.

Two paths:

1. text + optional title → ingest as a text note directly (no extraction needed)
2. files → extract text, dedup, index into OpenAI vector store

For files that need human review (image/audio/PDF), the frontend calls /extract first, lets the user edit the text, then re-submits via /ingest with text + source_type.