Here's a first draft of a shared Docker environment. The README includes some ideas about how to manage this stuff. In short, I think we should use standalone notebooks that create process step artifacts. Things that get created with ArcPy will not get run on the shared system but should still be able to be visualized/used by the rest of the system. By using ArcPy dependencies in small parts of the processing chain that can be run by those with a Windows and Arc License, we should be able to keep using Arc where needed while also moving toward a fully reproducible on any environment work flow.
The source project of this merge request has been removed.