Release v4.0.0 — LLM-judge coherence, structured labels, reproducible cases

Bump version to 4.0.0 integrating three merged PRs since v3.1.0:

- #4 LLM-judge coherence analysis (opt-in via --coherence-judge)
- #5 Enforce structured [Observed]/[Inferred] labels with parser fallback
- #6 Three reproducible case studies with live-API artifacts

Also fixes ruff E741 (ambiguous variable name) in labels.py to restore CI green.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>