feat(auth): add row/size caps + streaming to export-user-data

- Add per-collection row cap (default 10k, env EXPORT_ROW_CAP) via Prisma
  take on all findMany calls
- Add total size cap (default 100MB, env EXPORT_SIZE_CAP_MB); throws
  PayloadTooLargeException (413) when exceeded
- Convert response to Node.js Readable stream piped via NestJS StreamableFile
  to avoid large in-memory buffers
- Export ExportUserDataResult interface (stream + truncated flag) from handler
- Update controller to set Content-Type/Content-Disposition headers and
  return StreamableFile
- Document EXPORT_ROW_CAP and EXPORT_SIZE_CAP_MB env vars in Swagger
- Extend tests: row-cap assertion (take arg), size-cap 413 path, stream assertions

Fixes GOO-223 (M-1 from GOO-200 audit).

Co-Authored-By: Paperclip <noreply@paperclip.ing>

apps/api/docs/observability/README.md

@@ -0,0 +1,50 @@
+# Observability — Read-Model / Projector (RFC-003 Phase 0)
+
+Grafana dashboards and wiring notes for the read-model observability stack
+introduced in [GOO-192](/GOO/issues/GOO-192) under [GOO-94](/GOO/issues/GOO-94) §6 Phase 0.
+
+## Metrics
+
+All metrics live in the existing NestJS `metrics/` module
+(`apps/api/src/modules/metrics/`) and are scraped via the standard `/metrics`
+endpoint.
+
+| Metric                                  | Type      | Labels    | Purpose                                                   |
+| --------------------------------------- | --------- | --------- | --------------------------------------------------------- |
+| `read_model_projector_lag_seconds`      | Gauge     | `handler` | Seconds between latest source event and projector cursor. |
+| `read_model_refresh_duration_seconds`   | Histogram | `view`    | Duration of read-model / materialised view refreshes.     |
+| `read_model_reconciliation_drift_total` | Counter   | `model`   | Count of drift discrepancies found during reconciliation. |
+
+### Emit points
+
+Inject `MetricsService` and call:
+
+```ts
+metrics.setProjectorLag(handler, lagSeconds);
+metrics.recordReadModelRefresh(view, durationSeconds);
+metrics.recordReconciliationDrift(model, count?);
+```
+
+## Dashboard
+
+- File: `read-models-dashboard.json` (Grafana schema v38).
+- Import into Grafana (`Dashboards → Import → Upload JSON`), pick the Prometheus
+  data source.
+- Variables: `handler`, `view`, `model` — derived from Prometheus label values.
+- Panels:
+  1. Projector lag by handler (time series + thresholded)
+  2. Max projector lag (stat, RAG 30s / 120s)
+  3. Refresh duration p50/p95 by view
+  4. Refresh throughput (refreshes/sec) by view
+  5. Reconciliation drift rate by model (15m rate)
+  6. Total drift events in last 24h (stat, RAG 1 / 10)
+
+## Local verification
+
+```bash
+pnpm --filter @goodgo/api dev
+curl -s http://localhost:3001/metrics | grep read_model_
+```
+
+All three metric families should appear with `# HELP` / `# TYPE` headers even
+before any samples are recorded.