hack: create-kind-cluster.sh cannot make a dual-stack or IPv6-only cluster

[ygao-g]

Sub-task of #246, per bowei's concrete steps: "PR to add starting up Kind in dual-stack mode, installing Substrate on the cluster."

hack/create-kind-cluster.sh writes no networking: block, so every local cluster is IPv4-only and there is nowhere to reproduce the defects #246 lists. Fixes for them cannot be tested before they are proposed, which is what that thread has stalled on.

Goal

KIND_IP_FAMILY selects ipv4 (the default, byte-for-byte the current config), dual, or ipv6, and the resulting cluster is checkable rather than assumed:

• node InternalIPs
• the kubernetes ClusterIP
• status.podIP / podIPs[0]
• pod-to-pod connectivity
• the address family of the CoreDNS answer
• an image pull through the local registry (the node resolves kind-registry over the kind Docker network — a different path from the host's localhost:5001)

Two failure modes produce a cluster that looks fine and then dies minutes later with pods stuck at ContainerCreating. Both should be caught at creation rather than diagnosed:

• a kind Docker network left over from a previous ipv4 run and silently reused, because kind delete cluster leaves it alone while the registry container is still attached
• a Docker daemon with IPv6 disabled, which hands kind a v4-only network whatever it asked for

Non-goals

• No CI matrix cell. Per BenTheElder and bowei's reply, this is for local testing: IPv6 kind has upstream gaps and per-PR Actions time is scarce on the free runners.
• No Substrate component fixes. This only makes the cluster available; the Envoy V4_ONLY resolver, the 0.0.0.0 binds and the hard-coded actor interior network stay on [Bug] Data plane is hard IPv4 #246.