Add cross-binary integration for OSM ingest → API HTTP read

[koinsaari]

Summary

• Adds one integration test that drives a real OSM ingest, then reads the resulting place over real HTTP through the full middleware stack
• Catches drift in the shared persistence shape between cmd/ingestion and cmd/api (GORM tags, JSONB scanners, response converters) that single-binary integration tests cannot
• Pinned POI matches cmd/ingestion/main_integration_test.go — any fixture or transformer change flushes both tests in the same PR

Test plan

• `go vet -tags integration ./...` clean
• `golangci-lint run --build-tags integration ./cmd/api/...` clean (only pre-existing issues in places_integration_test.go remain)
• `go test -tags integration -timeout 300s ./cmd/api/...` — 65 passed
• `go test ./...` — 254 passed

🤖 Generated with Claude Code

[claude]

Claude finished @koinsaari's task in 2m 36s —— View job

---

Code Review

• Gather context (PR metadata, changed files)
• Read REVIEW.md for review instructions
• Read the new test file
• Read related test files for comparison
• Review the implementation
• Post review findings
• Set commit status (skipped — 403, integration lacks statuses:write)

---

No blocking issues. The test correctly catches persistence-shape drift between the ingestion and API binaries by writing via place.UpsertBatch (the ingestion path) and reading via a live HTTP request through the full middleware stack. All core invariants pass.

What the test exercises

The structure is sound:

1. runOSMIngestForCrossBinary directly calls the shared internal/ packages (place.UpsertBatch, identity.Sweeper) — the same code paths cmd/ingestion uses at runtime.
2. handlerForServer (with auth middleware, not handlerNoAuth) stands up the full cmd/api stack — consistent with other GET-endpoint tests such as TestHandleGetPlace_ReturnsPlaceWithAccessibility.
3. The pinned osm_id=521143390 assertion matches cmd/ingestion/main_integration_test.go's pinnedPOIs[0], so any transformer or fixture change flushes both tests together.

🟡 Nits

1. Sweep error should carry a diagnostic message (cmd/api/cross_binary_integration_test.go:70)

// current
if _, err := sweeper.Sweep(ctx, touched); err != nil {
    t.Logf("sweep: %v", err)
}

A sweep panic would surface as a confusing failure elsewhere. A short comment explaining why this is non-fatal (// sweep is opportunistic; the place is already persisted) would help future readers. Alternatively, t.Logf("sweep non-fatal (place already persisted): %v", err) makes the intent clear.

2. batchSize = 1000 is re-declared rather than imported (cmd/api/cross_binary_integration_test.go:35)

The canonical batch size lives in cmd/ingestion. A diverging value (e.g. tuned for memory) won't propagate here. The constant doesn't affect test correctness, but a comment noting it is intentionally decoupled would prevent a future reader from wondering if it needs to match.

3. Lat/lng are not verified after HTTP decode (cmd/api/cross_binary_integration_test.go:99–136)

Coordinates come through the geometry JSONB column too, and a broken Geometry.Scan would silently produce (0, 0). A two-line check would close that gap without needing exact values:

if got.Lat == 0 && got.Lng == 0 {
    t.Errorf("lat/lng both zero — geometry may not have scanned correctly")
}

4. Pluck("id", &touched) collects all DB IDs (cmd/api/cross_binary_integration_test.go:59–61)

After a fresh truncate + ingest this is equivalent to collecting RETURNING id values from UpsertBatch, so it's correct. A brief comment would prevent a reader from wondering if it's accidentally sweeping stale rows (// all IDs after fresh ingest equal the touched set).

Plus one similar item: the pinned OSM node ID 521143390 is hardcoded at line 84 rather than via a named constant — minor, and the PR description documents the coupling.

---