Release v0.7.0: evidence-aware, reliable reuse decisions

[five0nit]

Summary

Prepare Brief2Ship v0.7.0 for the approved GitHub release.

• Fail closed on failed/partial/empty discovery, insufficient inspection, and decision-critical unknowns; inconclusive CLI outcomes write receipts and return exit 5.
• Evaluate the full candidate set before display truncation; --limit cannot change the disposition or discard evaluated/inspection evidence.
• Preserve core intent through bounded deterministic query planning, relevance eligibility, and confidence-aware ranking.
• Preserve unknown dependency data and package-vs-repository evidence scopes.
• Separate MIT body recognition from authoritative license normalization; free-form notices require review.
• Add v2 receipts, --summary, PowerShell onboarding, and automated benchmark/type/lint/wheel/sdist gates.
• Finalize versioned v0.7.0 install routes while keeping historical demos explicitly pinned to v0.6.2.
• Correct a maintainer-local path in the final verification document; the existing public-document regression gate remains unchanged.
• Repair Windows raw-byte and resolved-path test expectations, add LF/CRLF coverage, and locate installed console scripts using the interpreter's sysconfig scheme rather than assuming proximity to python.exe.

Verification

• Fresh release validation: 193 tests passing and 24/24 synthetic decision-regression cases.
• Ruff critical selectors: passed; Pyright: 0 errors, 0 warnings.
• Prior implementation matrix: Python 3.11, 3.12, 3.13 and native Windows; independent quality re-review approved.
• Rebuilt wheel/source distributions passed metadata, checksum, source-payload comparison, extracted-source tests, and clean installed console smoke.
• Hosted CI passed on the release head, including Python 3.11/3.12/3.13 on Linux and Windows plus the optional Trafilatura adapter. Main CI remains required before tagging.

Migration and limits

Read docs/code-discovery.md before updating automation. brief2ship-discovery-v2 and brief2ship-score-v2 are explicit contract migrations. Exit 5 is not clean-build permission. Provisional reuse remains subject to required checks.

The 24 benchmark fixtures are synthetic regression controls, not live-provider or human-reviewed accuracy measurements. Ordinary copyright-bearing MIT files also require review under the conservative rule.

No new runtime dependencies, unbounded searches, or default candidate execution. This approval covers GitHub publication only; do not dispatch the separate PyPI publisher or modify the shared installed CLI.