Fix readability sentence boundary counting

[puneetdixit200]

Summary

• count sentence-ending punctuation without splitting inside decimals, URLs, or common abbreviations
• add regression coverage for Dr./Mr., decimal values, and domain names
• clamp cosine similarity scores into the valid [-1, 1] range to avoid float32 rounding leaking values above 1.0

Closes #15

Verification

• UV_CACHE_DIR=/tmp/assayer-15-uv-cache UV_PYTHON_INSTALL_DIR=/tmp/assayer-15-uv-python UV_PROJECT_ENVIRONMENT=/tmp/assayer-15-uv-venv uv run --python 3.12 --with pytest --with pytest-asyncio --with numpy --with ruff pytest -m "not integration" -q
• UV_CACHE_DIR=/tmp/assayer-15-uv-cache UV_PYTHON_INSTALL_DIR=/tmp/assayer-15-uv-python UV_PROJECT_ENVIRONMENT=/tmp/assayer-15-uv-venv uv run --python 3.12 --with pytest --with pytest-asyncio --with numpy --with ruff ruff check .
• git diff --check HEAD~1 HEAD