gke-deploy: get deployed objects from manifests

[Haihan-Jiang]

What changed

Change gke-deploy's readiness polling to fetch deployed objects with kubectl get -f - using the original manifest, instead of fetching by only kind and name.

Why

When a cluster has multiple CRDs with the same Kind in different API groups, kubectl get <kind> <name> can resolve the wrong resource. This caused gke-deploy to wait on ingressroutes.traefik.containo.us even when the applied manifest declared apiVersion: traefik.io/v1alpha1.

Using the manifest for the get operation preserves the resource's declared apiVersion and kind, matching the object that was applied.

Fixes #962.

Validation

• cd gke-deploy && go test ./core/cluster ./deployer ./services ./testservices
• cd gke-deploy && go test ./...
• Added a regression test with an IngressRoute manifest using apiVersion: traefik.io/v1alpha1 and a test kubectl response that must be consumed through the manifest-based get path.