hack: the ipv6 kubeconfig repoint breaks kubectl on a native-Linux Docker host #1099

hack/create-kind-cluster.sh:148-160 rewrites kind's https://[::1]:PORT kubeconfig entry to
https://localhost:PORT whenever IP_FAMILY=ipv6, on the reasoning in the comment that localhost
"lets the client pick a family that works from either side". It cannot: on Debian-family hosts
/etc/hosts gives ::1 the names ip6-localhost and ip6-loopback, not localhost, so
localhost resolves v4-only while the apiserver is published on v6 only.

Measured on Ubuntu 25.10 with a native Docker daemon: kind publishes 6443/tcp -> [::1]:42311,
https://[::1]:42311/healthz answers ok, and https://localhost:42311/healthz is refused. The
repoint replaces a working address with a broken one and every later kubectl in the script dies at
connect. It is only needed when the daemon lives in a VM (Lima on macOS), which is the topology
every IPv6-only run in this project has used so far — hence it went unnoticed. Introduced in #877;
#958 makes the repoint conditional on [::1] failing to answer, and this issue tracks landing that
on main if #958 does not.