feat(outcomes): learn accepted and rejected PR patterns by repo

[JSONbored]

Parent: #32

Goal

Make recommendations reflect what maintainers actually merge or close.

Requirements

• Analyze merged, closed-unmerged, stale, and open PRs by repo.
• Extract outcome patterns by path, language, labels, size, linked issue presence, test evidence, review churn, and author role.
• Separate maintainer-lane activity from outside-contributor evidence.
• Feed success/failure patterns into decision packs and agent next actions.

Acceptance Criteria

• Agent can say “you tend to merge well in this repo/path/type” or “this pattern has high closure risk.”
• Maintainer-owned repo activity does not inflate miner-lane evidence.
• Closed PRs and stale open PRs affect risk reasoning.
• Recommendations remain deterministic.

Testing Requirements

This issue must preserve the repo's CI gate:

• npm run test:ci must pass.
• Global coverage must remain at or above 95% for lines, statements, functions, and branches.
• Add or update tests for every new branch, fallback path, sanitizer rule, and regression.
• Add invariant/property-style tests when behavior depends on sorting, gating, public/private boundaries, scoring, queue pressure, or source-upload safety.
• Public GitHub output must be tested against forbidden language: wallet, hotkey, raw trust score, payout, reward estimate, farming, private reviewability, and public score estimate.

Required Test Scenarios

• Merged PR pattern fixture.
• Closed PR pattern fixture.
• Maintainer-lane fixture.
• Stale open PR fixture.
• Deterministic sorting regression.