Reject malformed corpus labels

[heznpc]

Summary

• validate corpus label entries before applying them
• reject non-object labels, non-string notes, invalid decisions, and duplicate label fingerprints as usage errors
• add regression tests for malformed object labels and duplicate array labels

Debug evidence

• malformed private manifest labels could previously be interpreted as unreviewed labels or fall through as less clear internal failures
• invalid label shape now exits 2 with a direct validation error
• real corpus output remains stable: 64 documents, 29 findings, 0 reviewed, 29 unreviewed, 0 stale labels

Verification

• npm test
• npm run validate:corpus
• node scripts/validate-corpus.mjs --manifest /tmp/rulemeter-real-corpus/manifest.json --format json --out /tmp/rulemeter-real-corpus/result.label-shape.no-text.json
• node scripts/validate-corpus.mjs --manifest /tmp/rulemeter-real-corpus/manifest.json --format json --strict --out /tmp/rulemeter-real-corpus/result.label-shape.strict.json exits 1 as expected
• npm run dogfood
• npm run pack:check
• npm run smoke:install
• npm audit --audit-level=high