A company runs a large containerized microservices platform on Amazon EKS across three accounts. Their existing observability stack uses self-managed Prometheus servers running on EC2 for metrics and a self-managed Grafana instance for dashboards. As the platform has grown, the Prometheus servers frequently run out of memory during high-cardinality query bursts, local storage fills up requiring manual intervention, and the operations team spends significant effort patching and scaling these instances. The company wants to reduce operational overhead, retain long-term metric storage, and preserve their existing PromQL queries and Grafana dashboards with minimal rework. Which approach BEST meets these requirements?