docs(models): GPT-5.6 Sol supersedes GPT-5.5 as cross-model reviewer

[BaseInfinity]

Summary

• OpenAI shipped a new flagship family (gpt-5.6-sol / gpt-5.6-terra / gpt-5.6-luna) superseding GPT-5.5, verified against primary OpenAI docs. Updates the cross-model-reviewer guidance every consumer repo inherits from this one.
• 6 live-guidance files updated to GPT-5.6 Sol (fallback: Terra) at xhigh: AI_SETUP_LANES.md, README.md, CLAUDE_CODE_SDLC_WIZARD.md, both SKILL.md copies, AGENTS.md, CLAUDE.md. Historical citations (README.md:151, CLAUDE_CODE_SDLC_WIZARD.md:113) intentionally untouched — regression-tested to stay that way.
• Adds a documented (non-default) escalation to max/Pro mode for unusually risky PRs in AI_SETUP_LANES.md's Final Review Policy.
• 8 new per-location regression tests in test-doc-consistency.sh, mutation-verified against 4 distinct swap scenarios (Sol→Terra, Terra→Luna, gpt-5.6-sol→gpt-5.6-terra, and a GPT-5.6 Sol→GPT-5.7 Sol version-bump probe) — not just an aggregate "mentions 5.6 somewhere" check.

Review history

• Fable (fresh instance): clean, one nit fixed (stale comment claiming a bash nameref where the code actually uses echo + command substitution — bash 3.x/macOS has no namerefs).
• Codex (GPT-5.6 Sol, xhigh) — 3 rounds to CERTIFIED:
  • Round 1: 2 P1s found and fixed — (a) the new escalation paragraph made 2 pre-existing lines stale (both still called xhigh "the highest" tier); (b) the new tests only checked for the 5.6 substring, passing under a demonstrated Sol→Terra or Terra→Luna swap.
  • Round 2: found the round-1 test fix still passed a GPT-5.6 Sol → GPT-5.7 Sol mutation (checked for Sol but not 5.6 together) — fixed by requiring both together.
  • Round 3: CERTIFIED, no further findings.

Test plan

• tests/test-doc-consistency.sh: 67/67, RED confirmed before edits, GREEN after
• Mutation-tested against 4 swap scenarios, all caught precisely at the mutated line
• Full test suite sweep before/after — no regressions (only the known pre-existing test-persist-score-history.sh fixture fragility, unrelated to this change, already tracked separately)
• Fable review (clean, 1 nit fixed)
• Codex xhigh (Sol) cross-model review, 3 rounds to CERTIFIED