feat(analytics): classify recommendation outcomes

[JSONbored]

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

Goal

Classify persisted recommendation outcomes into deterministic quality categories for later analytics.

Requirements

• Classify accepted, rejected, ignored, stale, merged, closed, and improved outcomes.
• Separate explicit user feedback from inferred lifecycle outcomes.
• Keep classification private and advisory.

Acceptance criteria

• Outcome categories are deterministic and test-covered.
• Inferred outcomes are marked as inferred, not treated like explicit user feedback.
• Public output avoids private scoreability, reward/risk, wallet, hotkey, raw trust, and private reviewability.
• This is not public scoring or autonomous GitHub action.

Validation expected

• Classification fixtures for explicit and inferred outcomes.
• Public/private sanitizer tests.
• npm run test:ci.