Greenfield implementation of a new pipeline system as successor to pypipegraph2.
It's got a bunch of good ideas, and somewhat working impls (sandboxing, web based watcher, decent tracebacks), and I think it shows that the 'build-and-discover' workflow with multiple stores can work - even though we don't have truly remote stores yet.
But the whole model isn't defined rigorously, and the ux from the 'write some code side' is awful (views/outputs/exports, one output symlink folder for everything, no way to mix multiple scripts etc).
I need to think some more about this before attempting this again.