[drk-ram]

Summary

• keep README license matches on one line or in an explicit Markdown license section
• normalize recognized SPDX identifiers while rejecting prose-only mentions
• prioritize structured card metadata and repository license files over README guesses
• add regressions for prose over-capture, section parsing, and source precedence

Validation

• .venv\Scripts\python.exe -m pytest -q tests\test_readme_license_extraction.py — passed (5 tests)
• .venv\Scripts\python.exe -m pytest -q — passed (full suite)
• regression tests against origin/main — failed as expected (5 failures)

Fixes #128