seems like doing round-robin within the cluster would be a sane default if there's no special load-balancing policy defined by SpiceDB?