- Add Rust reimplementation in rust/ for benchmarking (~5x faster, 100% match)
- Use underthesea character normalization (NFC + character map) in
src/normalizer.py instead of custom unicodedata logic
- Simplify src/parser.py by removing unused _classify_part function
- Add benchmark.py comparing Python vs Rust on all 10,602 ward records
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>