fix(signals): include recent merged PRs in outcome analysis

[philluiz2323]

Closes #312.

Problem

buildRepoOutcomePatterns built its analyzed PR set from only args.pullRequests and used recentMergedPullRequests purely as a by-number enrichment lookup. But the loader reads pullRequests (the pull_requests table) and recentMergedPullRequests (the separate recent_merged_pull_requests table) independently, and the backfill writes open PRs to pull_requests (marking vanished ones closed with no mergedAt) while merged PRs go to the separate table. So a repo's historical merged PRs were absent from the analysis: outsideContributorMergeRate collapsed toward 0, falsely firing the "Outside contributor PRs rarely merge here / high closure risk" pattern and suppressing "merge well here", and every outcome dimension was blind to merged history. This feeds buildContributorStrategy, repo-fit, reward-risk, and the decision pack, so healthy repos were reported as high-closure-risk.

Fix

• New normalizeRecentMergedOutcome helper converts a recent_merged_pull_requests record into a decided/merged RepoOutcomePullRequest, deriving maintainer-lane / author-role from the stored payload's author_association (the table has no association column).
• The analyzed set is now the union of pull_requests rows and merged records that exist only in recent_merged_pull_requests (deduped by PR number, requiring a real mergedAt).
• Reconciliation fix: a PR with a genuine recent-merged record (carrying mergedAt) is counted as merged, not closed_unmerged, even when the open-PR reconciliation only saw it vanish. A recent-merged enrichment record with mergedAt: null does not flip a PR to merged.

Tests

Added a fail-on-revert test: with only closed shells in pullRequests and 9 merged PRs living only in recent_merged_pull_requests, the result must report totals.merged: 9 and the "Outside contributors merge well here" success pattern, not the false high-closure-risk pattern. The existing recent-merged tests only used merged PRs that also appear in pullRequests, so they never exercised the merged-only path.

vitest run repo-outcome-patterns (+service) 30/30; consumers + e2e (signals, signals-v2, decision-pack, queue, integration/api) 143/143; tsc --noEmit clean.

[JSONbored]

@philluiz2323 the fix looks useful and CI is green, but the PR title needs to follow the repo title convention.

Required change:

• Retitle this to something like fix(signals): include recent merged PRs in outcome analysis.

Validation expected:

• Keep npm run test:ci green after any update.

[philluiz2323]

@philluiz2323 the fix looks useful and CI is green, but the PR title needs to follow the repo title convention.

Required change:

• Retitle this to something like fix(signals): include recent merged PRs in outcome analysis.

Validation expected:

• Keep npm run test:ci green after any update.

done

[JSONbored]

@philluiz2323 this still needs cleanup before it can be reviewed.

A few notes:

• The problem area is valid.
• The title has a trailing period and does not meet this repo's PR-title policy. (fixed)
• The PR body is missing the current template evidence.

Required changes:

• None for now.

Validation expected:

• Focused repo outcome pattern tests.
• Full CI green after any rebase.