What steps should you take to address high read latency in a global chat application using a multi-regional Cloud Spanner instance after a recent deployment caused degraded performance?