This PR fixes the regression introduced by PR #175

I initially switched the kubectl commands from apply to create as apply was idempotent and safer when running commands in a retry loop. However, the Installation CRD of the Calico Operator is so large that (1.1 MB) that it exceeds the kubernetes annotation size of 262144 bytes (every resource created via kubectl apply has the kubectl.kubernetes.io/last-applied-configuration annotation added to it)

Since the holodeck e2e flakes happen when executing the following command in a retry loop, it should suffice to just kubectl apply the calico custom-resources.yaml file

with_retry 3 10s kubectl --kubeconfig $KUBECONFIG apply -f https://raw.githubusercontent.com/projectcalico/calico/${CALICO_VERSION}/manifests/custom-resources.yaml