[hatemadi15]

Closes #247.

Add ccc duplicates to find cosine-similarity groups in an existing SQLite vector index. It uses stored embeddings without running the daemon or embedding model, visits each unordered pair once in bounded-size blocks, and groups threshold matches transitively. Supports text and JSON output, plus language, path, and minimum-line filters. The exact scan takes quadratic time in the number of selected chunks.

Validation

• 44 focused tests passed, including four CLI tests against an on-disk sqlite-vec index. They cover filters, transitive matches across comparison blocks, text and JSON output, missing indexes, and NaN input.
• Ruff lint/format, focused mypy, and git diff --check passed.
• Broader run: 310 passed, 27 failed, 17 errors, 8 deselected. The workspace blocks daemon sockets, also reproduced on main. Full test-tree mypy reports the same existing errors as main.