Eventually it would be nice for the classic parser to be removed entirely or retooled to produce an intermediate DOM based on parsoid's model

That seems like an epic on its own.

Yes, I'm on an epic task filing kick. ;)

A strawman proposal for this can be found at T114454: [RFC] Visual Templates: Authoring templates with Visual Editor.

I believe it would be best to implement a more rigorous separation of code, data, and styling at the same time.

In so doing, you'd avoid exposing many of the details of the parser's representation to modules, which should help with the "generating intermediate DOM from the classic parser" part of the task.

I am not familiar with the new parser at all, so feel free to ignore this if it is irrelevant...:

In any case it MUST be distiguishable, what kind of replaced content is being there (currently it IS recognizible by the tag name within the strip marker).
In other words - it MUST be possible to find out whether the replaced content is <ref>, <nowiki>, <score>, whatever...
Yet better it would be, if both the original source of the tag and its result were available as well. (Artificial example to explain: <ref>Foo</ref> vs. <ref>{{PassValue|Foo}}</ref> - both behave same (Template:PassValue just returns whatever is given as an argument) - but as seen, the original code within <ref></ref> is different.)

In T133543#2237551, @Danny_B wrote:

and its result

This is unlikely as a general thing due to issues like T63268 and T73167.