How should connection resiliency and transient retries be designed for production in Entity Framework Core?

Retry only transient errors, keep operations idempotent where possible, add observability, and avoid long-lived manual connections.