fix(copyright): filter trade-secret notice false positives

[mstykow]

Summary

• add a refiner junk-pattern rule for extracted copyright and trade secret[s] boilerplate fragments
• add targeted copyright regressions for the refiner and detector pipeline, using an anonymized Acme Inc. notice sample
• preserve the real copyright/holder extraction while preventing the confidentiality-notice false positive

Scope and exclusions

• Included:
  • refiner-layer junk filtering for trade-secret boilerplate fragments
  • unit and detector-level regression coverage for the reported false positive
• Explicit exclusions:
  • broader detector/tree-walk refactors toward ScanCode's parser/token architecture
  • golden fixture changes or output-schema changes

Intentional differences from Python

• none intended; this change narrows Provenant's behavior toward ScanCode by dropping a false-positive boilerplate fragment that ScanCode does not emit

Follow-up work

• Created or intentionally deferred:
  • evaluate whether a small amount of earlier candidate/token filtering would reduce future late-stage junk-pattern additions without adopting ScanCode's architecture wholesale