feat: publish durable queue and spool gauges

cmd/rvbox-server/main.go

@@ -109,6 +109,7 @@ func run(configPath string) error {
 	}
 	recoveryContext, cancelRecovery := context.WithCancel(context.Background())
 	defer cancelRecovery()
+	go publishServerTelemetry(recoveryContext, persistence, health)
 	// Recovery is deliberately asynchronous: liveness and incident inspection
 	// remain available while committed-range checks run.  Readiness becomes
 	// true only after the real SQLite/segment recovery completes.
@@ -239,3 +240,24 @@ func run(configPath string) error {
 		return httpErr
 	}
 }
+
+func publishServerTelemetry(ctx context.Context, persistence *store.Store, health *observability.Health) {
+	ticker := time.NewTicker(10 * time.Second)
+	defer ticker.Stop()
+	for {
+		sampleContext, cancel := context.WithTimeout(ctx, time.Second)
+		telemetry, err := persistence.Telemetry(sampleContext)
+		cancel()
+		if err != nil {
+			health.Inc("telemetry_read_failure")
+		} else {
+			health.SetGauge("queue_depth", float64(telemetry.QueuedCommands))
+			health.SetGauge("server_charged_bytes", float64(telemetry.ChargedBytes))
+		}
+		select {
+		case <-ctx.Done():
+			return
+		case <-ticker.C:
+		}
+	}
+}