Fix kubectl-in-curlimages/curl: use DNS name instead of command substitution

[Copilot]

curlimages/curl doesn't ship kubectl, so $(kubectl get svc my-app-svc -o jsonpath="{.spec.clusterIP}") inside that pod silently fails. The canary exercise also needs to run the curl from inside the cluster since ClusterIP isn't reachable from the host.

Changes

• c.pod_design.md: Replace the broken host-shell while/curl/kubectl get svc pattern with a self-contained throwaway pod that curls the service by DNS name:

# Before
while sleep 0.1; do curl $(kubectl get svc my-app-svc -o jsonpath="{.spec.clusterIP}"); done

# After
kubectl run curl --image=curlimages/curl --rm -it --restart=Never -- \
  sh -c 'while sleep 0.1; do curl my-app-svc; done'

Using the in-cluster DNS name (my-app-svc) eliminates the need for kubectl inside the pod entirely.