LGTM, I have a few nits below. I also think the namespace code could have space in the abstract base class and that encode/decode implementations are empty but that could be addressed later if we want.

I have a few small comments, and one larger comment for later consideration, but overall this looks great. Having this coherent structure will be a really big win for maintainability even before we potentially start extracting more shared logic.

I also really, really like having the checked in output files for this kind of PR. Being able to see the (trivial) output changes makes it really easy to have confidence that nothing got accidentally changed in the reorganizing.