fix(benchmark): verify apache hadoop target

[mstykow]

Summary

• verify apache/hadoop at dbcc7cd797100e6b32cd84f85b53a5193a5f9af0 with a benchmark-grade compare-outputs run and record the result in docs/BENCHMARKS.md
• fix generic sourcemap, copyright, metadata-cleanup, tree-walk, and comment-author paths exposed by Hadoop parity review
• regenerate docs/scan-duration-vs-files.svg after adding the Hadoop timing row (278.63s vs 4575.96s, 16.42× faster)

Issues

• Covers: Hadoop benchmark verification for https://github.com/apache/hadoop at dbcc7cd797100e6b32cd84f85b53a5193a5f9af0
• Closes:

Scope and exclusions

• Included:
  • generic sourcemap license/party handling improvements
  • generic copyright/holder extraction and metadata-cleanup fixes exposed by Hadoop
  • Hadoop benchmark entry and regenerated benchmark chart
• Explicit exclusions:
  • no parser .expected.json or golden fixture updates
  • no attempt to chase reviewed JDiff/CHANGELOG/NOTICE noise that did not justify another general fix

Intentional differences from Python

• keep reviewed differences where Provenant is deliberately cleaner or more specific, including de-duplicated repeated MIT sourcemap detections and rejecting ScanCode's Apache-2.0 AND LGPL-2.1-only over-assertion on plain Apache notice headers

Follow-up work

• Created or intentionally deferred:
  • deferred low-value NOTICE/author cleanup that did not justify another robust generic fix after the Hadoop review

Validation

• cargo test test_collapse_repeated_sourcemap_license_detections_combines_concrete_detections
• cargo test test_detect_multiline_x_editable_actual_line5_shape_survives_postprocess_boundary
• cargo test test_detect_multiline_x_editable_actual_line5_shape_end_to_end
• cargo test test_collect_trailing_orphan_tokens_absorbs_name_before_legal_tail
• cargo test test_derive_holder_from_simple_copyright_string_strips_and_onwards_prefix
• cargo test test_extract_comment_author_supplements_handles_html_comment_by_line
• npm run check:docs
• cargo run --manifest-path xtask/Cargo.toml --bin compare-outputs -- --repo-url https://github.com/apache/hadoop.git --repo-ref 7b92f0bc7f8fb8f243635493da76e8f59475ef87 --profile common

Benchmark evidence

• Target snapshot: apache/hadoop @ dbcc7cd797100e6b32cd84f85b53a5193a5f9af0
• Compare artifacts: .provenant/compare-runs/20260506T074747Z-hadoop-28100
• Files: 16,370
• Timing: Provenant 278.63s; ScanCode 4575.96s (16.42× faster)

Expected-output fixture changes

• Files changed: None
• Why the new expected output is correct: no golden or .expected fixtures changed; this branch updates runtime behavior and benchmark docs only