ANG-007: Markdown preprocessing & embedding pipeline

[yugalkaushik]

• Joplin now indexes notes into chunks and stores their vectors locally. This PR adds a provider that reads those vectors directly from that index. It checks if the index is ready, then pages through all stored vectors, averages them into one vector per note, and normalizes them. It handles model changes mid-flight by restarting, and notes that aren't indexed yet get reported as errors without blocking everything else. No preprocessing on our side since Joplin already handles cleanup and chunking during indexing.
• Updated the joplin's api
• Added export as SVG