Rich text: a clear split between the core and the implementation for block context

[ellatrix]

This refactoring is internally to rich text and not affecting public APIs.

We’d like to use core rich text without implementation specific things outside block context at some point.

I already started this separation when I created the rich-text package, but this separation was not complete as there was still much implementation specific logic left in the core rich text component. For example, undo, split, merge, enter etc. handling is very block context specific. Core rich text still relies on a block client ID, which is of course block specific.

For web, core rich text will become a simple behavioural hook returning a ref callback. I'm not sure how a native core rich text component (or maybe hook returning props and children) could look like, but the core rich text for web and native will look different, which means the implementation of it will look a bit different too.