test(windows): expand native writer, CLI and package contract coverage #34

Windows native transport is already implemented (#4/#10), but its automated verification is narrower than the macOS/Linux contract jobs. At TS 0.1.0 (0edc4f8), windows-native runs Node 22/24 install, build/type-check through test:windows, two Windows tests, and audit. Those tests cover Claude named-pipe authentication/submission (including stale PID evidence) and an unlocked lock probe. They do not exercise a held Codex writer, guarded queue dispatch, CLI/SSH outcome boundaries or installed-package smoke on Windows.

This is a platform verification gap, not a claim that Windows functionality is absent or that the recorded live Windows ACKs did not occur. The later README-only merge 2e8d111 does not change this coverage.

Completion criteria

• Add isolated Windows held-lock/native-owner fixtures covering stable identity, wrong/competing/changing owners, lock replacement and pre-submit revalidation. Exercise actual Windows locking/process inspection rather than treating POSIX results as Windows evidence.
• Exercise guarded Codex CLI queue dispatch: dry-run submits nothing, one valid send submits once, invalid/inactive/ambiguous evidence refuses, and native failure/timeout after dispatch retains an unknown outcome without retry. Keep optional future inactive/wake behavior outside this baseline suite.
• Run Windows CLI/SSH framing and result-contract checks, including implementation/version preflight, Windows .cmd/PowerShell paths, JSON stdin, refused versus unknown outcomes and secret-free errors. Use isolated fake SSH/native endpoints, not production sessions.
• Adapt package smoke for Windows launcher and permission semantics; test artifact contents, supported native dependency loading, version/empty discovery, typed import, clean isolated install and uninstall on Node 22/24. Do not execute a POSIX npm shim as JavaScript or use executable-bit checks as proof of Windows launchability.
• Make the new checks required in the existing release gate. Diagnose fixture/process-readiness timeouts explicitly rather than masking failures with retries or removing assertions.
• Update docs: maintain Python parity matrix and npm migration guide #33 / VALIDATION evidence by OS, architecture, Node/Codex version and date. Distinguish deterministic fixture results, native submission and separately observed TUI ACK; do not infer Windows arm64 or live Node 22 coverage from Windows x64/Node 24 observations.

Dependencies and priority

Priority P1, target milestone 0.1.1 — Validation and documentation. This can proceed against current 0.1.0 independently of #16/#17. Coordinate new scenario fixtures as those features land; use #33 for the evidence matrix. No new production messages, platform claims or publication are authorized by this test issue.

Evidence and review

Identified by the Python-maintainer session review STS-PARITY-REVIEW-20260927-A and rechecked against source by the TS session.

• Windows CI job
• Windows tests
• Current package smoke
• Writer implementation
• Dated live evidence and limits