chore(autoreview): sync canonical review skill

[vincentkoc]

What Problem This Solves

The repo-local autoreview skill had drifted from the OpenClaw organization source, including older model defaults and review behavior.

Why This Change Was Made

Sync .agents/skills/autoreview exactly from openclaw/agent-skills@4664d27da471d1cb71bebdd9845dc8a6c56d6bbe.

• defaults Codex review to gpt-5.6-sol with high reasoning
• retries gpt-5.6-terra only when the account cannot access Sol
• adds AGENTS.md plus CLAUDE.md -> AGENTS.md directing future changes to the organization source first
• aligns the complete skill directory to prevent downstream drift

User Impact

No acpx runtime behavior changes. Maintainers get the current organization-wide review workflow.

Evidence

• exact subtree parity: 0c6893e09512920c2af2ddb71c3a5de308cd47f0
• canonical chore(deps): bump commander from 13.1.0 to 14.0.3 #82 Sol/high review clean
• 26 CLI tests passed
• 192 hardening tests passed with 1 expected skip; the isolated C# performance check passed in 1.53s
• git diff --check passed
• native formatting was previously green; the final rerun was unavailable because this worktree intentionally has no local node_modules

AI-assisted: yes. The synchronized source, tests, and review findings were manually verified.