feat: cross-document synthesis — query across all notebooks

[vikranthreddimasu]

Summary

The killer feature. Researchers can now ask questions that synthesize information across ALL their notebooks at once.

Cross-Document Synthesis

• Click "All" in the chat header to switch to cross-notebook mode
• Questions retrieve and rank results from every notebook's ChromaDB collection, merged by L2 distance
• The LLM prompt explicitly instructs the model to compare, contrast, and name disagreements across sources
• Sources in the panel show which notebook each chunk came from (sage-colored notebook label)
• Click "Single" to switch back to querying just the active notebook

Backend

• query_across_notebooks in vector_store.py: queries multiple ChromaDB collections, merges results, re-ranks by distance, returns top_k
• prepare_prompt_cross_notebook in rag.py: builds a prompt that groups excerpts by notebook + document and instructs the LLM to synthesize across all sources
• stream_reply in chat.py: routes to cross-notebook RAG when notebook_ids list has 2+ entries

Frontend

• crossNotebookMode in Zustand store with "All/Single" toggle in chat header
• useChat.ts sends notebook_ids (all notebook IDs) instead of notebook_id when in cross mode
• SourcePanel.tsx shows notebook name labels in sage uppercase text when cross-mode is active

Test Coverage

6 new tests for multi-collection query: merge results, top_k limit, empty notebooks skipped, single notebook fallback, all-empty returns empty, results sorted by distance.

Tests: 7 → 8 files (+1 new). 28 passed, 1 skipped.

Pre-Landing Review

No issues found. Backend changes follow existing RAG patterns. No new auth surface (single-user desktop app).

Test plan

• All backend tests pass (28 passed, 1 skipped)
• Manual: Create 2+ notebooks with documents, click "All", ask a cross-cutting question
• Manual: Verify source panel shows notebook labels for each source
• Manual: Click "Single" to return to single-notebook mode

🤖 Generated with Claude Code