fix(backfill): hydrate merged PR changed files in segment backfill

[galuis116]

Closes #352.

Summary

The production segment-pipeline backfill stored changedFiles: [] for every recent-merged PR (backfillRecentMergedSegment hard-coded []), while only the legacy monolithic path fetched the real list. upsertRecentMergedPullRequest also unconditionally overwrote changedFilesJson, clobbering any populated list, and no detail-sync hydrates merged-PR files. So recent_merged_pull_requests.changedFiles was always empty in production, blinding the outcome path/size/test-evidence dimensions and collision file-overlap (both read mergedDetail.changedFiles).

Scope

• src/github/backfill.ts — fetch each merged PR's changed files in the recent-merged segment via fetchPullRequestFiles (mirrors the monolithic path; reuses the existing mapWithConcurrency, REST -> GraphQL fallback, and rate-limit handling).
• src/db/repositories.ts — upsertRecentMergedPullRequest keeps the existing changedFilesJson on conflict when the incoming list is empty (no clobber).
• test/unit/backfill.test.ts — segment-hydration + coalesce tests.

No schema, API, or public-surface changes.

Validation

npx vitest run test/unit/backfill.test.ts                                   # 53/53
npx vitest run test/unit/queue.test.ts test/unit/data-spine.test.ts \
  test/unit/data-quality.test.ts test/integration/api.test.ts \
  test/integration/routes-errors.test.ts                                    # 103/103
npx tsc --noEmit                                                            # clean
git diff --check                                                            # clean (no whitespace errors)

Branch coverage stays >= 97% (both coalesce ternary branches and the segment path are covered by the new tests). CI validate is green.

Safety

• The only data added is each PR's changed-file paths, fetched from the same /pulls/{number}/files GitHub endpoint the monolithic backfill already uses — no secrets, no diffs, no file content, and no new external surface or privileges.
• File fetches use the existing installation/public token and the existing rate-limit-aware fetch path, and fall back to [] on failure (a warning is recorded, no throw).
• The coalesce guard only prevents data loss — an empty incoming list never erases a populated one — and cannot introduce stale data, because a non-empty incoming list still overwrites.

Notes

• The extra per-PR file fetches in the segment match what the monolithic path already does; concurrency is unchanged (mapWithConcurrency(..., 8)).
• Distinct from [Bug]: buildRepoOutcomePatterns ignores the recent_merged_pull_requests table — merge-rate analysis runs almost without merged PRs, falsely flagging healthy repos as "high closure risk" #312 / fix(signals): include recent_merged_pull_requests in outcome pattern #316 (which fix the analysis ignoring recent-merged) — this fixes the backfill data those consumers read.

[JSONbored]

@galuis116 the backfill problem statement is useful, but the PR needs hygiene fixes before merge.

A few notes:

• Hydrating changed files for recent merged PRs is a real data-quality improvement.
• Preserving non-empty existing file lists on re-upsert is the right invariant.
• The current title/body do not meet this repo's PR requirements.

Required changes:

• Retitle to fix(backfill): hydrate merged PR changed files in segment backfill.
• Update the PR body to the current template with Summary, Scope, Validation, Safety, and Notes.
• Include exact validation commands and security/privacy notes for the backfill data surface.

Validation expected:

• Keep the current green CI.
• Include the targeted backfill tests plus git diff --check in the PR body.

[galuis116]

Thanks for the review — addressed the hygiene items (no code changes from the reviewed diff):

• Retitled to fix(backfill): hydrate merged PR changed files in segment backfill (PR + commit).
• Rewrote the body to the template: Summary, Scope, Validation, Safety, Notes.
• Added exact validation commands, the targeted backfill tests, git diff --check (clean), and security/privacy notes for the backfill data surface (only changed-file paths from the same /pulls/{n}/files endpoint the monolithic path already uses; no new surface).

CI is green and re-running on the amended commit. Ready for another look.

[JSONbored]

@galuis116 this is in good shape from this pass.

A few notes:

• The segment backfill now hydrates changed-file paths instead of storing empty arrays.
• The coalesce guard avoids clobbering existing file data with empty incoming lists.
• The validation evidence is specific and covers the affected backfill, queue, data-quality, API, typecheck, and diff-check surfaces.

No code changes are requested from this review pass.