Fix the architecture of In-Publication Mode

Summary: Currently, In-Publication mode is kind of a horrible hack. Before we get to Experiment Mode, we need to completely rewrite this.

The issue is about how and where we process Create and Modify/ChangeProps. Currently, this is all smooshed into SpaceCore, but it really doesn't belong there -- the result is that SpaceCore is dependent on the InPublicationActor, which is evil and wrong. This needs to be more decoupled.

The heart of the problem is illustrated by modifyThing() -- really, that's the delicate bit. The problem is that, in order to do its thing for a Publishable Instance, it needs both the Model (which lives in Core) and the Instance (which lives in InPublication). Indeed, deciding whether a Change/Modify request is Publishable or not requires the Instance (which knows the Model) and the Model (which actually contains the flag).

This just barely works for InPublication -- we are squishing that into Core. But Experiments will break the camel's back: we need a deep rewrite for those.

The key observation is that the decision points about creating and modifying -- the permission-checking and so on -- fundamentally depend on knowing the entire State of the Space as this User sees it. Which means that those decisions need to happen in the UserSpaceSession, not in SpaceCore.

So the necessary plan is to lift all of that machinery out of SpaceCore, probably into a new trait, then move that trait over to OldUserSpaceSession. The Session will make all the decisions based on the enhancedState, and then send the Events in more or less raw form to the Core, the InPublication or the currently-active Experiment, depending on the change in question. (Note that Publication and Experiments need to be able to co-exist -- think about what this should look like.)

Complicating all of this is the Space plugins. Look at each of these, and how they fit into the equation. This needs a lot of deep design before we act on it.