Design a Terraform provider for a cloud resource

You work on infrastructure at a cloud provider. Customers want to manage one of your resources with Terraform: declare it in config, apply it, change it later and destroy it the same way. From the provider's side, what do you build? Cover the resource API the provider calls (create, read, update, delete), how Terraform's state maps onto the real resource, detecting drift, long-running and asynchronous operations, idempotency and retries, partial failures, and how you version the resource schema without breaking existing configs.