fix(decision-council): enforce evidence-based challenge changes

[9uiLe]

Summary

• add a deterministic challenge audit guard for recommendation changes
• retain round-one recommendations for persuasion-only or unsupported switches
• permit evidence-backed and concrete reasoning/option corrections
• distinguish protocol cleanliness from safety to synthesize
• detect two-way and multi-advisor cross-switches and reject convergence claims
• preserve confidence and accepted-concern updates independently from position changes

Decision Council dogfood

• froze O1 persuasion resistance, O2 legitimate correction, and O3 cross-switch classification
• used fresh, explicitly verified Fable 5 and Codex gpt-5.6-sol/high sessions
• both advisors identified deterministic synthesis enforcement—not more prompt text—as the remaining gap
• resolved the blocking-vs-neutralization disagreement through challenge: Fable changed on OVERLOOKED_EVIDENCE from Issue [Decision Council] Challenge can reverse recommendations without new evidence #53's preserve-prior requirement; Codex retained neutralization with explicit safety/integrity signals

Testing

• node --test plugins/second-opinion/tests/*.test.mjs
• python ~/.codex/skills/.system/skill-creator/scripts/quick_validate.py plugins/second-opinion/skills/decision-council
• ./scripts/verify-versions.sh
• git diff --check

Closes #53