fix(copyright): generalize compare-target cleanup

[mstykow]

Summary

• generalize shared copyright, holder, and author cleanup across detector/refiner/scanner paths so compare-target noise is reduced without introducing Flutter-specific string blacklists
• add supporting extraction and normalization work for storyboard/XML text= copyright evidence, image metadata dedupe, ISO-date/unknown-year holder recovery, templated URL cleanup, escaped SPDX tag normalization, and a final cleanup pass for regex/template junk, generated-resource email noise, and distribution-metadata author tails
• validate the branch with focused regressions, copyright golden + library-profile checks, a documented Flutter benchmark checkpoint, a clean pnpm compare check, and fresh exact compare reruns with ScanCode cache hits

Issues

• Covers: flutter/flutter residual compare cleanup, pnpm/pnpm regression safety, and broader cross-target copyright/holder/author false positives
• Closes:

Scope and exclusions

• Included: shared detector/refiner/scanner cleanup, metadata extraction refinements, finder/SPDX normalization fixes, benchmark documentation refresh, two copyright golden expectation updates, and compare-run verification evidence
• Explicit exclusions: no claim that flutter/flutter is fully clean yet; remaining sampled deltas still include planet.frag normalization duplicates, the Dart project authors vs the Dart project, Flutter code-fragment leftovers like Copyright 0 absl::StrCat(errors 0 ) / Copyright void, and scancode-toolkit fixture-token leftovers like AUTH AUTHS 2730, COMPANY 1411, and MAINT 26382

Intentional differences from Python

• Provenant continues to preserve structured contributor/author evidence from committed AUTHORS files and labeled image metadata when that evidence is present, even where ScanCode often leaves those surfaces empty

Follow-up work

• Created or intentionally deferred: if we continue, the next high-value buckets are the remaining code-fragment junk in Flutter license-checker fixture text and the remaining uppercase fixture-token author noise in scancode-toolkit src/cluecode/copyrights.py

Expected-output fixture changes

• Files changed: testdata/copyright-golden/copyrights/misco4/linux-copyrights/Documentation/bpf/bpf_devel_QA.rst.yml, testdata/copyright-golden/copyrights/licco.txt.yml
• Why the new expected output is correct: the old expectations kept stale false-positive author forms; after the generalized cleanup the focused regressions, copyright golden, and library-profile validation all pass with the corrected expectations

Compare artifacts

• Flutter benchmark/checkpoint recorded in docs: .provenant/compare-runs/20260427T223247Z-flutter-8526/
• pnpm clean validation checkpoint: .provenant/compare-runs/20260428T124247Z-pnpm-1113/
• exact rerun at pushed HEAD before the final mini-pass: .provenant/compare-runs/20260428T135306Z-scancode-toolkit-92561/ (ScanCode cache hit)
• exact rerun at pushed HEAD before the final mini-pass: .provenant/compare-runs/20260428T140858Z-camel-12496/ (ScanCode cache hit)
• latest dirty-tree rerun including the final mini-pass: .provenant/compare-runs/20260428T151953Z-flutter-6927/ (ScanCode cache hit)
• latest dirty-tree rerun including the final mini-pass: .provenant/compare-runs/20260428T152754Z-scancode-toolkit-16046/ (ScanCode cache hit)