unstaple v0.1.0

First release.

Splits a multi-document scanned PDF into separate, sensibly-named PDFs using deterministic text-layer signals: page-number resets, letterhead (header fingerprint) shifts, date discontinuities, and blank-page separators. Includes --dry-run with per-cut confidence and reasons.

v0.1 scope: text-layer PDFs only (image-only scans need OCR first; native OCR support is on the roadmap). Validated end to end against synthetic stapled fixtures; 69 tests passing.