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

`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.