chore: make just recipes work on Windows

[tonyfettes]

Windows recipes currently depend on POSIX shell behavior and launchers, so normal just commands fail before reaching the underlying tools. This change runs all four justfiles with Windows PowerShell on Windows while preserving the default shell on macOS/Linux.

• Resolve editor paths with absolute_path, preserve paths containing spaces, and omit an empty optional --browse-root argument.
• Invoke Playwright through Node and select npm.cmd on Windows.
• Select python on Windows for the existing Python lifecycle test, with a PYTHON override.
• Add test-cram and select Git for Windows' Bash explicitly, avoiding the WSL launcher on PATH. Keep the root test steps as ordered dependencies so overrides propagate.
• Document Python, Git Bash, path overrides, and the NSIS packaging prerequisite; remove the stale jq prerequisite.

Windows validation

Executed real recipes using Windows PowerShell, just 1.37.0, the repository-pinned MoonBit v0.10.12+1634b282e in an isolated toolchain, the VS 2022 amd64 environment, Node 23.11.0, and Python 3.11.9.

Passed:

• just build (native and JS), just editor-build, just cef-setup, and just desktop-build-scripts-check.
• just test-cram: 24/24; just viz-test-browser: 13/13.
• just desktop-package: ZIP and installer generated after adding the installed NSIS directory to the process PATH. The installer was not run.
• Supplementary JS checks and tests: strict check passed, 3217/3217 tests passed.
• Inspector and editor serve/dev startup returned HTTP 200, including editor paths containing spaces. Desktop dev created a window and reported renderer bridge ready; native UI interaction was not tested. All validation services were stopped.
• moon info, moon fmt, moon fmt --check, and git diff --check; no generated interface changes.

The full Windows suite is not green:

• just check: native --deny-warn rejects five unused fields/variables in bgjobs and worktree code.
• just test and just test-moon: native deepseek/client diagnostics tests hung and were manually interrupted. Later root test stages were therefore exercised separately.
• just test-turn-finish: the run case passes; the serve process does not exit after agent_finished and is terminated by the 120-second test timer.
• just editor-test: JS 1875/1875; Wasm 1086/1090 and native 1220/1224, with the same four diagnostics-refresh failures; wasm-gc has no test entry.
• just editor-test-browser: 105/107; the component-only rerun is 68/70, reproducing the same read_api/viewer_api geometry failures.
• just desktop-test-browser: 56/57; the transcript Markdown local-file-link locator fails.

These failures remain unresolved in this PR; they have not been compared against a separate base-branch run. Tests that self-skip for missing online credentials or sh are not counted as coverage. Validation above is on Windows; macOS/Linux were not rerun.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-09T05:00:22.522496Z	67f89f4	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.