start: deploy model from registry

[dberenbaum]

Report

We have the sagemaker example where we deploy, but we don't expose the endpoint nor provide a way for the user to try deployment themselves.

Some options to explore (we could also try to partner with whoever we choose here):

• could start with just building a docker image and pushing it to dockerhub
• https://github.com/replicate/cog with or without hosting on https://replicate.com/
• flyio like in https://iterative.ai/blog/mlem-nanogpt-modal-flyio