feat: add native Reasonix skill installation

[esengine]

Reasonix is not an install target, and the generic skill dispatches Task(description=...) rather than Reasonix's native task(prompt=...). This adds explicit Reasonix installation and generated skill bundles for POSIX and Windows PowerShell.

• Add graphify install --platform reasonix and graphify reasonix install/uninstall, with --project support.
• Resolve global skills via REASONIX_HOME, ~/.reasonix on macOS/Linux, and APPDATA/reasonix on Windows, with the roaming-profile fallback.
• Preserve project/user scope and existing AGENTS.md guidance. Native subcommand installs register AGENTS.md; runtime hooks are not configured.
• Use task prompt/write_paths and include the extraction prompt; fall back to in-session extraction if task is unavailable.
• Generate both skill bundles and expected fixtures from skillgen sources. Include both bodies in the wheel and test their packaging explicitly.

Closes #1381.

Validation executed on Windows:

• New CLI regression tests fail 5/5 on upstream HEAD; pass 5/5 with this implementation.
• Relevant install, version, generic-agent, and wheel-packaging suites: 177 passed.
• Full ruff check: passed. Focused pyright using the venv interpreter: zero errors.
• skillgen --check, both new-host --audit-coverage checks and --schema-singleton: passed.
• Built the wheel; packaging tests cover both Reasonix shell variants and their reference bundles.
• graphify update . completed using local AST extraction.

Validation limits:

• The broad Windows pytest run completed with 6,380 passed, 146 failed, and 176 skipped. It is not claimed green. One newly collected Reasonix roundtrip case exposed missing test isolation for APPDATA; its fixture was corrected, and both Reasonix roundtrip cases now pass (in addition to the 177 focused checks). Four existing Gemini/Claude refresh and uninstall failures reproduce on an unchanged upstream worktree. The remaining broad failures have not all been audited.
• Full pyright reports 639 errors with base dependencies, including absent optional dependencies and existing test typing issues; focused checks above pass.
• Linux/macOS host selection is mocked. Real provider authentication and live semantic extraction through Reasonix are not claimed.

Most added lines are generated skill/reference artifacts and matching expected fixtures, following this repository's skillgen workflow. Implementation prepared with OpenAI Codex; commit includes co-author disclosure.

[github-actions]

Thanks for the pull request, @esengine. A maintainer will review it soon.

Want to talk it through while it is in review? Come join us on our Discord server. For longer-form discussion there is also GitHub Discussions.

A couple of things that speed up review: make sure the test suite passes on Python 3.10 and 3.13, and that the change keeps extraction deterministic.