fix(copyright): clean up Boost markup regressions

[mstykow]

Summary

• clean up shared copyright/author detection regressions exposed by the boostorg/boost compare-outputs --profile common run
• add durable copyright golden regressions for XML author attributes, DocBook authorgroup extraction, and Boost CSS selector noise, and update stale golden expectations that were treating file names/scripts as authors
• record the verified boostorg/boost result in the C++ parser verification scorecard

Scope and exclusions

• Included:
  • markup author extraction and entity decoding improvements in the shared copyright detector
  • junk filtering for CSS/prose/path-like author and holder noise
  • focused detector/refiner tests, narrow copyright golden suite coverage, docs scorecard update, and repeated Boost compare validation
• Explicit exclusions:
  • no parser-family feature expansion beyond the shared detection fixes surfaced by the compare run
  • no changes to the kept Provenant-better normalization/improvement deltas unless they were clear regressions

Intentional differences from Python

• keep cleaner Provenant-better results from the Boost compare run, including the extra real Boost copyright/author detections and cleaner normalization/deduplication in markup-heavy files
• drop bogus golden “authors” that were actually file names or generator scripts (DynamicClockGatingTable.ctb, EnableASIC_StaticPwrMgtTable.ctb, EnableDispPowerGatingTable.ctb, createinit.py)

Expected-output fixture changes

• Files changed: testdata/copyright-golden/authors/boost_xml_author_attr_entities.xml.yml, testdata/copyright-golden/authors/boost_docbook_authorgroup.html.yml, testdata/copyright-golden/copyrights/boostbook_css_noise.css.yml, testdata/copyright-golden/copyrights/misco4/linux-copyrights/drivers/gpu/drm/amd/include/atombios.h.yml, testdata/copyright-golden/copyrights/misco4/linux-copyrights/drivers/gpu/drm/radeon/atombios.h.yml, testdata/copyright-golden/copyrights/misco4/linux-copyrights/drivers/media/usb/dvb-usb/af9005-script.h.yml
• Why the new expected output is correct:
  • the new Boost fixtures lock in the compare-run regressions this branch fixed
  • the updated Linux fixture expectations now match the improved shared author filter, which correctly treats those prior “authors” as code/path noise rather than people