You should be able to tell Tilt about your cluster if Tilt can't infer things

Right now Tilt uses some pretty simple heuristics to decide what kind of Kubernetes cluster you have:

1. Strict matching on the name of the cluster, like with minikube.
2. Fuzzy matching on the name of the cluster, like with GKE
3. Path to kubeconfig, like with k3d.
4. Fuzzy matching of path to kubeconfig, like with KIND.

Tilt uses this information to decide whether it can use an internal image registry, among other things. It's important for the Tilt user experience that we use an internal image registry whenever possible since that speeds up image push times.

Unfortunately it's easy for a user to inadvertently break Tilt's cluster inference code by renaming clusters or moving kube configs.

We should provide some sort of escape hatch to tell Tilt to consider cluster named X to be minikube/microk8s or something. I'm not entirely sure yet. I'm opening this issue to spur discussion.

Start by reading the cluster-detection heuristics in internal/k8s/env.go, including the name and kubeconfig path matching described in the issue. Trace how the inferred cluster type affects internal image registry selection. The work is done when users have a documented escape hatch for overriding inference and the supported cases are verified.