e2e: reach clusterip clusters from the external test harness #923

Follow-up from #920.

Problem

The e2e suite runs the operator in the pytest process (kopf_runner), outside the test cluster, with CRATEDB_OPERATOR_TESTING=true. In that mode get_host connects through the data service's LB address, or externalDNS for traefik. A clusterip cluster has neither, so get_host raises TemporaryError until the test ends. The clusterip e2e tests therefore use wait_for_healthy=False and never connect: bootstrap, health and suspend/resume are not tested end to end for clusterip.

Production is not affected. Outside test mode get_host always returns the headless crate-discovery-<name>.<namespace>, and #920 has a unit test that pins this for every exposure.

Options

• Port-forward (preferred): in test mode, forward a local port to a CrateDB pod through the Kubernetes API. get_host returns only a host and connection_factory always uses port 5432, so the port must be passed through as well. No test-only code in operator handlers.
• Test-only LB service: the test creates an extra LoadBalancer service with the data-service selector, and test-mode get_host uses it for clusterip. Cheap, but adds a test-only path to operator code.

On minikube, minikube tunnel also routes the service CIDR, but that does not help the AKS nightly job.