feat(agent): feed aggregate outcome quality into confidence

[JSONbored]

Parent roadmap: #127
Parent phase: #237
Parent epic: #278
Related advanced phase: #269
Related base feedback issue: #96

Goal

Use private aggregate outcome quality to adjust future recommendation confidence without making public reward or score claims.

Requirements

• Feed aggregate quality signals into deterministic confidence/ranking logic.
• Keep the signal advisory and explainable.
• Do not expose private aggregate quality as public trust, payout, or score prediction.

Acceptance criteria

• Recommendation confidence can incorporate aggregate outcome quality.
• Confidence changes are explainable and bounded.
• Public summaries avoid private scoreability, reward/risk, wallet, hotkey, raw trust, and private reviewability.
• This is not public scoring or autonomous GitHub action.

Validation expected

• Ranking/confidence fixtures with positive, negative, sparse, and missing outcome quality.
• Public/private sanitizer tests.
• npm run test:ci.