backend: GDPR export + retention cleanups + worker metrics (BE-1/2/3)

BE-3 observability: expose the worker's Prometheus metrics on :6060/metrics
(apns/fcm/asynq histograms + a new cache_ops_total counter were recorded all
along but never scraped — which is why those dashboard panels read empty); add
the worker containerPort, the vmagent worker scrape job, and two additive
NetworkPolicies. Instrument cache Get/Set hit/miss.

BE-2 retention: three periodic Asynq cleanup crons mirroring the reminder-log
cleanup — notifications (90d), webhook dedup log (180d), audit_log (365d).

BE-1 GDPR data export: POST /api/auth/export/ enqueues a low-priority Asynq job
that gathers all of the user's data (owned residences + their tasks/contractors/
documents/share-codes, plus profile/notifications/prefs/push-tokens/subscription/
audit log), zips one JSON file per category, and emails it as an attachment.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

cmd/worker/main.go

@@ -232,7 +232,11 @@ func main() {
 	mux.HandleFunc(jobs.TypeOnboardingEmails, jobHandler.HandleOnboardingEmails)
 	mux.HandleFunc(jobs.TypeReminderLogCleanup, jobHandler.HandleReminderLogCleanup)
 	mux.HandleFunc(jobs.TypeUploadCleanup, jobHandler.HandleUploadCleanup)
+	mux.HandleFunc(jobs.TypeNotificationCleanup, jobHandler.HandleNotificationCleanup)
+	mux.HandleFunc(jobs.TypeWebhookLogCleanup, jobHandler.HandleWebhookLogCleanup)
+	mux.HandleFunc(jobs.TypeAuditLogCleanup, jobHandler.HandleAuditLogCleanup)
 	mux.HandleFunc(worker.TypeTaskCompletedNotification, jobHandler.HandleTaskCompletedNotification)
+	mux.HandleFunc(worker.TypeDataExport, jobHandler.HandleDataExport)
 
 	// Register email job handlers (welcome, verification, password reset, password changed)
 	if emailService != nil {
@@ -281,6 +285,23 @@ func main() {
 	}
 	log.Info().Str("cron", "30 * * * *").Msg("Registered pending_uploads cleanup job (runs hourly)")
 
+	// Data-retention cleanups (BE-2). Staggered off the 3:00 reminder cleanup to
+	// avoid piling DELETEs onto the same Neon connection window.
+	if _, err := scheduler.Register("0 2 * * *", asynq.NewTask(jobs.TypeNotificationCleanup, nil)); err != nil {
+		log.Fatal().Err(err).Msg("Failed to register notification cleanup job")
+	}
+	log.Info().Str("cron", "0 2 * * *").Msg("Registered notification cleanup job (daily 02:00 UTC, 90d retention)")
+
+	if _, err := scheduler.Register("30 2 * * 0", asynq.NewTask(jobs.TypeWebhookLogCleanup, nil)); err != nil {
+		log.Fatal().Err(err).Msg("Failed to register webhook log cleanup job")
+	}
+	log.Info().Str("cron", "30 2 * * 0").Msg("Registered webhook log cleanup job (weekly Sun 02:30 UTC, 180d retention)")
+
+	if _, err := scheduler.Register("30 3 * * 0", asynq.NewTask(jobs.TypeAuditLogCleanup, nil)); err != nil {
+		log.Fatal().Err(err).Msg("Failed to register audit log cleanup job")
+	}
+	log.Info().Str("cron", "30 3 * * 0").Msg("Registered audit log cleanup job (weekly Sun 03:30 UTC, 365d retention)")
+
 	// Handle graceful shutdown
 	quit := make(chan os.Signal, 1)
 	signal.Notify(quit, syscall.SIGINT, syscall.SIGTERM)
@@ -292,6 +313,12 @@ func main() {
 		w.WriteHeader(http.StatusOK)
 		_, _ = w.Write([]byte(`{"status":"ok"}`))
 	})
+	// Expose Prometheus metrics so vmagent can scrape the worker. The
+	// apns_send_*, fcm_send_*, asynq_job_* and cache_ops_* series have been
+	// recorded on this process all along — they were just never exposed, which
+	// is why those dashboard panels read empty. Same :6060 as health; in-cluster
+	// only (not externally published).
+	healthMux.Handle("/metrics", prom.HTTPHandler())
 	healthSrv := &http.Server{
 		Addr:              workerHealthAddr,
 		Handler:           healthMux,