fix: harden matching correctness

[SerDrozd]

Summary

Fix several correctness issues in matching and runtime version reporting.

Changes

• derive the runtime version from package metadata instead of duplicating it in application code
• preserve Unicode letters during text normalization
• parse Excel-style datetime values correctly during date matching
• support common US and European amount formats
• preserve accounting-style negative amounts
• keep the documented base matching weights when optional rules are omitted, then normalize the active rules
• add regression coverage for the corrected cases

Validation

• uv run ruff check .
• uv run mypy src tests
• uv run pytest — 50 passed
• uv build
• git diff --check