feat!: replace ChromaDB with sqlite-vec, extend CLI

[ernestkoe]

Summary

• Replace ChromaDB with sqlite-vec for fully local, telemetry-free vector storage (~200KB, pure C, no transitive deps like posthog/grpc/opentelemetry)
• Replace custom chunk_by_heading() with Chonkie RecursiveChunker using markdown-aware splitting rules (headings > paragraphs > lines > sentences)
• Add CLI commands: similar, context, --path-filter on index
• Add obsidian-rag short alias alongside obsidian-notes-rag
• Fix similar/context note lookup with direct SQL query (get_by_file) instead of broken KNN+filter pattern

Details

sqlite-vec migration

• Two-table schema: chunks (metadata) + chunks_vec (vec0 virtual table for KNN)
• Contract tests written first, then backend swapped — same 11 tests validate both implementations
• Net dependency change: -1,417 lines in uv.lock, 63 packages removed

Chonkie chunker

• Inline markdown rules (no huggingface_hub dependency)
• Lazy singleton chunker instance for performance
• Splits by: h1 > h2 > h3 > h4 > paragraphs > lines > sentences > words

CLI extensions

• obsidian-rag search "query" — semantic search
• obsidian-rag similar "Path/To/Note.md" — find related notes
• obsidian-rag context "Path/To/Note.md" — show note + related context
• obsidian-rag index --path-filter "Daily Notes/" — selective re-indexing

Test plan

• 25 tests passing (test_store.py, test_indexer.py, test_cli.py)
• Smoke tested: uv tool install --force, re-indexed 1203 files / 2162 chunks
• Verified search, similar, context, stats, index --path-filter commands
• Verified MCP server get_similar and get_note_context tools updated

🤖 Generated with Claude Code