fix(copyright): tighten wrapper and notice extraction

[mstykow]

Summary

• Verify PX4/eigen @ 7cf1c0179eb0f5499dfc1bffbd229783a7865fe1 with compare-outputs --profile common, record the reviewed benchmark result in docs/BENCHMARKS.md, and refresh docs/scan-duration-vs-files.svg.
• Tighten shared copyright extraction so Provenant keeps meaningful raw notice text while stripping wrapper shells such as PRODUCT_COPYRIGHT, applicationLegalese, LegalCopyright, and storyboard text= wrappers from emitted copyrights.
• Reduce shared false positives and boundary bleed across detector/refiner/scanner layers for generated-doc junk, table/test labels, minpack example prose, locale timestamp suffixes, and large bundled notice/license files, with focused regression coverage plus rerun compare validations on Flutter and Eigen.

Issues

• Covers:
• Closes:

Scope and exclusions

• Included:
  • Compare artifacts: .provenant/compare-runs/20260504T211702Z-eigen-62479/, .provenant/compare-runs/20260505T102504Z-flutter-12526/, and .provenant/compare-runs/20260505T104455Z-eigen-34962/
  • Focused regression tests in src/copyright/refiner/tests.rs, src/copyright/detector/tests.rs, and src/scanner/process/copyright_test.rs
  • Compare-layer URL slash-only delta suppression from the earlier branch work in xtask/src/bin/compare_outputs.rs
  • Benchmark/chart refresh and the parity-safe correct-copyright-minpack golden update
• Explicit exclusions:
  • No parser/package-surface changes; PX4/eigen remains manifest-free under the shared profile
  • No compare-output policy change that hides raw copyright wrapper differences behind extra semantic normalization
  • No risky generic rewrite of malformed source-email tails; malformed Eigen <email lines remain source-faithful for now

Intentional differences from Python

• Preserve meaningful raw emitted copyright text such as All rights reserved. when it is part of the source notice, while keeping the normalized shadow/refined comparison semantics separate.
• Preserve source-faithful names and Unicode when Provenant is more specific than ScanCode, rather than collapsing output back to ScanCode-style lossy normalization.

Follow-up work

• Created or intentionally deferred:
  • Remaining large-bundle normalization and holder-boundary differences in Flutter sky_engine/LICENSE
  • Remaining Eigen naming/email-shape deltas where the source itself is malformed or Provenant is arguably more complete than ScanCode (for example Desire NUENTSA WAKAM)
  • Raw-output wording differences such as Copyright (c) Microsoft Corporation vs bare (c) Microsoft Corporation that are lower-priority than the false-positive and wrapper-shell fixes completed here

Expected-output fixture changes

• Files changed: testdata/copyright-golden/copyrights/misco3/correct-copyright-minpack.txt.yml
• Why the new expected output is correct: the detector now recovers the structured Copyright Notice (1999) University of Chicago form and matching holder directly from the copied fixture text, and the focused regression tests cover that behavior.