Fix edge-case bugs found in a thorough code review

Speed up knowledge graph builds by extracting pages concurrently

Harden knowledge graph build (isolation, status reset, embed fallback)

Fix knowledge graph embedding model discovery

Add knowledge graph search over extracted text

Support optional shared online-OCR key via env var

Disable local handwriting gracefully when packages absent

Restructure layout and refresh palette

Redesign UI with light and dark themes

Add PDF extractor app and Dockerfile

initial commit