Fix section hits across mismatched sources

[Abduhu1]

Summary

• Require a section hit to match both the evidence source_id and section heading.
• Prevent identical section headings from different guidelines from being counted as section hits.
• Add a regression test covering the mismatched-document case.

Why

Previously, _section_hit compared only section text. This meant evidence from nice-ng238 could receive a section hit against gold evidence from nice-ng136 when both happened to use the same heading, even though the document hit correctly failed.

Section-level retrieval should only count when the section belongs to the correct source document.

Tests

• uv run --frozen pytest -q — 56 passed
• uv run --frozen ruff check . — passed
• uv run --frozen ruff format --check . — passed
• git diff --check — passed