`form-js-viewer` refactor, splitting view and state management

[Skaiir]

What should we do?

We need to split the form-js-viewer state management logic from its rendering in two different projects.

Why should we do it?

There are many motivations for this change:

• Our reliance on preact to compute things like feel expressions, visibility, ect... is slowing us down more and more, many timing issues are created as a result of tethering render to computations and it also affects performance
• Our tests are unreliable and slow from their reliance on jsdom and waiting for UI event propagation
• Splitting the two would make implementing new renderers in potentially other frameworks / different toolkits rather simple
• Removing the viewer's dependency on the viewer would be a nice improvement as we currently do some quite hacky things with it

How we are going to do it?

• [x] Agree on form component & form structure interfaces
• [ ] https://github.com/bpmn-io/form-js/issues/1189
• [ ] https://github.com/bpmn-io/form-js/issues/1208
• [ ] Migration of custom components
• [ ] Integrate the new packages into the viewer
• [ ] Build out the editor "core" submodule (defines what the editor will supply the renderer, we don't need a full package for this, as it'll be small)
• [ ] Integrate the module + the renderer into the editor
• [ ] Consider integration concerns into other products
• [ ] Final checks

Internal discussions ongoing, see comments for the brainstorms.

[Skaiir]

We are currently considering to build two sub-packages to handle the separation. These would be independent from each-other which means that you could either use the core package with a custom renderer, or the renderer package with any type of schema corresponding to the visual contract we want to build up between the two packages.

I briefly sketched out how we see the core architecture's direction (early ideas anyways). But we could opt for a near-stateless transformation pipeline that would look something like this:

The advantage of a pipeline like this would be that it would simply eliminate side-effects, and make the forms truly deterministic relative to the only thing that can actually change within it (the managed form state).