fix(copyright): drop legal-prose false positives

[mstykow]

Summary

• filter generic legal-prose copyright and holder fragments in the shared copyright refiner instead of special-casing the local compare target
• normalize smart quotes in refiner punctuation stripping and add preservation coverage for legitimate notice-shaped copyrights and holders after narrowing the earlier over-broad refiner pass
• add detector and refiner regressions for the Meta SDK-style legal prose and update one to_improve golden fixture where bare copyright is now correctly treated as junk holder noise

Issues

• Covers:
• Closes:

Scope and exclusions

• Included:
  • shared refiner heuristics for copyright and holder junk filtering
  • preservation tests for (c) the European Community 2007 and Copyright (c) 1988, 1993
  • regression tests for the observed legal-prose false-positive class
  • one justified golden-fixture expectation update for misco4/to_improve/junk-copyright-224.txt.yml
  • rerun validation with cargo test --features golden-tests --test copyright_golden test_golden_copyrights and compare-outputs on .provenant/LICENSE using profile common
• Explicit exclusions:
  • the existing trailing-slash URL normalization parity difference between https://developers.meta.com/horizon and https://developers.meta.com/horizon/

Follow-up work

• Created or intentionally deferred:
  • investigate whether the remaining URL trailing-slash normalization mismatch should be normalized in Provenant output, compare reduction, or left as-is

Expected-output fixture changes

• Files changed: testdata/copyright-golden/copyrights/misco4/to_improve/junk-copyright-224.txt.yml
• Why the new expected output is correct: the fixture input is only copyright _copyright, so a bare copyright holder is junk noise rather than a real holder identity; the narrowed refiner now preserves legitimate copyright-prefixed notice holders while dropping this singleton junk case