fix(copyright): recover malformed copyright year ranges

[mstykow]

Summary

• treat malformed copyright year ranges like 20010-2011 and existing oversized tails like 2010-20224 as year-like during detector gating instead of dropping the detection
• wire the detector and tree-walk year checks through the shared helper so holder extraction still works when the malformed range is tokenized as a cardinal
• add unit and end-to-end regressions covering malformed first-year ranges without changing normal email-bearing copyright handling

Scope and exclusions

• Included: copyright detector token-utils year-like recognition, detector/tree-walk year gating, and targeted regression coverage
• Explicit exclusions: lexer year grammar changes, golden fixture updates, and broader normalization of arbitrary long numeric ranges

Intentional differences from Python

• Provenant now preserves malformed copyright ranges with obvious year-shape typos in more cases instead of requiring the first range token to be a strict four-digit year before extraction can proceed.