fix(copyright): stop directive absorption and support single-file compare targets

[mstykow]

Summary

• stop copyright detections from absorbing lowercase trailing directive comment lines such as @lint-ignore-every while preserving the full Confidential and proprietary notice text
• add focused detector/refiner regressions plus an anonymized copyright golden fixture for the reported case and the @History: author follow-up case
• fix xtask compare-outputs for single-file --target-path runs by staging file inputs consistently so ScanCode and Provenant compare the same logical path

Issues

• Covers: reported copyright false positive on trailing directive lines and the follow-up single-file compare-outputs failure
• Closes:

Scope and exclusions

• Included: copyright candidate grouping and same-line suffix handling, regression tests/golden coverage, and the xtask single-file compare workflow
• Explicit exclusions: broader heuristic tightening for other theoretical multiline marker classes beyond the reproduced cases

Intentional differences from Python

• For the reproduced two-line fixture, Provenant now drops the trailing directive line but preserves Confidential and proprietary; current ScanCode avoids the directive append too, but truncates the notice earlier at Confidential

Follow-up work

• Created or intentionally deferred: broader investigation of other weak multiline marker classes (for example URL-follow-up lines or author-like prose) was deferred because this branch closes the reproduced bugs without new failing fixtures

Expected-output fixture changes

• Files changed: testdata/copyright-golden/copyrights/copytest/anonymized_lint_directive_not_absorbed.txt and .yml
• Why the new expected output is correct: the trailing @lint-ignore-every line is scanner metadata rather than copyright text, while the full first-line notice Confidential and proprietary should remain part of the extracted copyright