fix(copyright): recover eigen benchmark author parity

[mstykow]

Summary

• keep comment-led Copyright (c) ... by ... lines on the shared holder-extraction path, stop treating Xerox Corporation as junk holder text, and tighten follow-up holder recovery so prose-only copyright notice of ... spans do not leak holders while real names like 42North Inc. are preserved
• rerun compare-outputs for ValveSoftware/eigen @ e9c43151265207fd3366bba21cddd61141ff402c with the maintained common profile and the same shared ScanCode cache identity
• refresh the eigen benchmark checkpoint and chart to the latest end-state snapshot (eigen-43257), which keeps 0 package deltas, 0 dependency deltas, and 42 vs 42 top-level license detections, then sync the small set of copyright-golden expectations whose cleaner holder/copyright output is now intentionally preserved

Issues

• Covers: benchmark verification for ValveSoftware/eigen
• Closes:

Scope and exclusions

• Included:
  • src/copyright/detector/pattern_extract/extraction/groups.rs
  • src/copyright/detector/postprocess_transforms/year_repairs.rs
  • src/copyright/detector/postprocess_transforms_test.rs
  • src/copyright/detector/tests_copyright_holder_pipeline.rs
  • src/copyright/detector/tests_false_positives.rs
  • src/copyright/detector/token_utils/filters.rs
  • src/copyright/refiner/holders_junk_patterns.rs
  • src/copyright/refiner/tests.rs
  • docs/BENCHMARKS.md
  • docs/benchmarks/scan-duration-vs-files.svg
  • targeted copyright golden expectations under testdata/copyright-golden/copyrights/
• Explicit exclusions:
  • no attempt to match ScanCode's weak Distributed holder overcapture in bench/eig33.cpp

Intentional differences from Python

• No intentional Python-compatibility divergence was introduced. The remaining bench/eig33.cpp holder delta is a weak license-prose overcapture (Distributed) that Provenant intentionally does not model as a holder.

Follow-up work

• Created or intentionally deferred:
  • none for this PR beyond any future review of unrelated legacy acknowledgment-shaped author/copyright differences surfaced by the eigen compare run

Expected-output fixture changes

• Files changed:
  • testdata/copyright-golden/copyrights/misco2/regexhq/regexhq-115.txt.yml
  • testdata/copyright-golden/copyrights/misco2/regexhq/regexhq-204.txt.yml
  • testdata/copyright-golden/copyrights/libopenthreads12-libopenthreads.copyright.yml
  • testdata/copyright-golden/copyrights/wxWindows_Library_.0_variant.yml
• Why the new expected output is correct:
  • markdown-link and ellipsis-tailed holder/copyright cleanup now preserves the cleaner short form instead of stale dangling ( or ... suffixes
  • follow-up false-positive fixes keep prose-only copyright notice of ... text from leaking holders and preserve real derived names like 42North Inc.
  • validated with:
    • cargo test test_refine_holder_keeps_xerox_corporation
    • cargo test test_derive_holder_from_simple_copyright_string_strips_by_prefix
    • cargo test test_derive_holder_from_simple_copyright_string_keeps_leading_digits
    • cargo test test_comment_led_by_keyword_holder_backfilled
    • cargo test test_copyright_notice_of_prose_does_not_emit_xerox_holder
    • cargo test --features golden-tests --test copyright_golden test_fixture_misco3_not_real_copyrights -- --nocapture
    • cargo test --features golden-tests --test copyright_golden test_golden_copyrights -- --nocapture
    • cargo test --features golden-tests --test copyright_golden test_golden_holders -- --nocapture
    • cargo run --manifest-path xtask/Cargo.toml --bin compare-outputs -- --repo-url https://github.com/ValveSoftware/eigen.git --repo-ref e9c43151265207fd3366bba21cddd61141ff402c --profile common