user-specified image push strategy

Currently, Tilt has a few different mechanisms for how to push images (* see below)

I was talking to the kubemacs team on k8s slack (https://github.com/kubemacs/kubemacs), and they have a more complicated cluster configuration.

They'd like to specify a custom command-line for loading images. Having Tilt auto-detect it from the kubeconfig is difficult, because their kubeconfig is already pretty complex.

Something like 'kind load', but configured manually

• current image push strategies:

• If if finds a "tilt.dev/registry=localhost:5000" annotation on the node, it will use that. This is what powers Kind local registries (https://github.com/windmilleng/kind-local)
• If it detects you're using kind, it will try to figure out how to use kind load to copy the image
• If it detects you're using microk8s' local registry plugin, it will try to push to the microk8s registry
• If it detects you're using minikube w/ docker, it will try to talk to the docker socket directly and build in-cluster
• if it detects you're using docker for desktop, it will assume it doesn't need to push
• if you specify a default registry with the Tiltfile default_registry() command, it will use that

Start by tracing Tilt's existing image-push strategies and the Tiltfile default_registry() entry point, comparing the kind load path with the other detection paths. The work is done when users can configure a custom image-loading command for a cluster while the existing automatic strategies continue to work.