fix(detection): improve apache camel compare parity

[mstykow]

Summary

• improve shared UTF-16 decoding so BOM-less or corrupted-BOM inputs stay on the decoded-text path, including the Camel template files that previously missed Apache-2.0
• recover Apache, Spring, and OpenShift notice attributions and trim written-by prose spillover in shared author heuristics, including the Camel SBOM XML false-positive sentence
• filter version-shaped SBOM pseudo-emails before truncation and record the explicit Apache Camel benchmark checkpoint in docs/BENCHMARKS.md

Issues

• Covers: Apache Camel compare target from docs/implementation-plans/package-detection/PARSER_VERIFICATION_SCORECARD.md
• Closes: none

Scope and exclusions

• Included: shared text decoding in src/utils/file.rs, shared copyright/author heuristics, shared email host filtering, Camel benchmark/chart refresh
• Explicit exclusions: unrelated remaining Camel compare deltas outside the three targeted gaps, such as the longstanding top-level package-count mismatch

Intentional differences from Python

• keep the UTF-16 improvement generic by recognizing BOM-less and corrupted-BOM UTF-16 shapes before falling back to Latin-1 or binary-string extraction
• prefer unique SBOM emails before the cap only after filtering version-shaped pseudo-hosts, instead of keeping obviously fake package-version addresses in the truncated result set

Follow-up work

• Created or intentionally deferred:
  • recorded compare artifacts:
    • baseline before these fixes: .provenant/compare-runs/20260426T074449Z-camel-39047
    • first fully-fixed benchmark checkpoint: .provenant/compare-runs/20260426T202057Z-camel-80585
    • immediate repeat for timing variance: .provenant/compare-runs/20260426T203456Z-camel-82481
  • the benchmark row uses camel-80585; the repeat camel-82481 was retained only to check fluctuation and came out 7.84s slower (1.67%), which suggests the earlier 473s vs 514s spread was mostly different code state rather than pure runtime noise
  • remaining non-targeted Camel compare deltas are left for later triage

Expected-output fixture changes

• Files changed: none
• Why the new expected output is correct: this branch changes shared scanner behavior and validates it with targeted Rust tests plus repeated compare-output artifacts; no checked-in golden or expected-output fixtures required updates