feat(signals): wire burden forecast through intelligence + MCP with freshness

[mkdev5]

Closes #22.

Summary

• New getBurdenForecast reader on the existing burden_forecasts table (src/db/repositories.ts).
• New shared services helper loadOrComputeBurdenForecastResponse (src/services/burden-forecast.ts) returns an envelope with generatedAt, ageSeconds, and a freshness: \"fresh\" | \"stale\" marker (6 h window matching DECISION_PACK_MAX_AGE_MS). Falls back to a computed forecast for known repos that have no cached snapshot yet, so callers see actionable data on first hit.
• buildRepoIntelligenceResponse (src/api/routes.ts) attaches burdenForecast + a burdenForecastFreshness slice (source, generatedAt, ageSeconds, freshness) to both snapshot and computed branches. Reads from the dedicated PK-indexed table, so request-time scans stay bounded.
• New MCP tool gittensory_get_burden_forecast (src/mcp/server.ts) for parity with the private intelligence surface; returns not_found for unknown repos.

Acceptance criteria → coverage

• AC1 "Large queue fixtures complete within Worker-safe bounds" — ragflow/sure-scale fixture (120 open PRs + 1 stale) at burden-forecast.test.ts:16-24 asserts level: \"critical\" and runs in unit-test time.
• AC2 "Forecasts identify top burden sources and cleanup actions" — duplicate-trend and stale-PR fixtures assert forecast.duplicateTrend and findings[*].code for queue_growth_risk + stale_review_load; the existing builder also emits finding.action for cleanup hints.
• AC3 "Repo intelligence includes freshness and data-quality warnings" — integration test at api.test.ts:723-735 asserts the intelligence response carries burdenForecast + burdenForecastFreshness: { source: \"snapshot\", freshness: \"stale\", ageSeconds: > 0 } when the cached row is older than 6 h.
• AC4 "No broad request-time scans are needed" — new spy regression in burden-forecast.test.ts:103-122 asserts listIssueSignalSample, listOpenPullRequests, and listRecentMergedPullRequests are NOT called when a cached forecast exists.

Validation

• npm run typecheck
• npx vitest run test/unit test/integration — 352 tests pass
• npx vitest run --coverage — branches at 95.07% (over the 95% gate)

Safety

• Backend-only change
• Private MCP/API surface only; no public-comment changes
• No new public-facing copy
• Reuses existing dedicated burden_forecasts table; no migration
• CHANGELOG.md unchanged (per CONTRIBUTING.md rule)

Open questions for review

• Standalone REST route: deliberately NOT added a GET /v1/repos/:owner/:repo/burden-forecast handler — test/unit/openapi.test.ts keeps the path on the "removed" list, matching the project pattern of exposing repo signals via the /intelligence aggregator + MCP tools rather than dedicated GET endpoints. Happy to add it if you'd rather expose it separately.
• Freshness threshold: defaulted to 6 h matching DECISION_PACK_MAX_AGE_MS. Open to a different value.
• Storage shape: kept the dedicated burden_forecasts table (one row per repo, primary-key access). Could move to signal_snapshots for consistency with other signals, but that's a refactor beyond this PR's scope.

Relationship to other in-flight work

• Independent of PR feat(signals): wire issue quality reports through snapshot + REST + MCP + opportunities #56 (issue-quality wiring) — different storage tables and different consumer paths. Both follow the same cache-or-compute helper pattern.

[mkdev5]

@JSONbored please review the PR when you have a chance and let me know your feedback.

[JSONbored]

#61 feat(signals): wire burden forecast through intelligence + MCP with freshness
action: merge
required:

• None obvious. Snapshot-first with bounded computed fallback and passing validate.

[mkdev5]

thank you for your review!