This is essentially a handmade caching system for your content.
Once you approach the problem that way, you can also consider other forms of caching, like having a dynamic, server-side generated site and a cache store sitting in front of it. That cache store can be some fancy caching system or it can be the output of a crawler than runs once a day against your dynamic site and outputs to a set of static html files that get served by your webserver.
Whatever I do has to apply incrementally as I (slowly) transfer everything into the new SSG controlled repository.
Normalised is definitely where I want to be for the canonical representation.
I guess there'd be two canonical representations actually; the SSG 'source' and the normalised hybrid HTML+SSI.
I hesitate because some of my existing content uses mod-include and other apache httpd functionality in different ways for different purposes.
I considered segmenting the site so some parts are cached/proxied using SSR (Server Side Rendering for those that wonder) but then I'm introducing artificial barriers that in the future I may forget are there and trip over whilst trying to throw some quick PoC or demo up.
So the SSG will understand and use the per-page metadata to generate index-type pages but actual inclusion of content is delayed until it is requested from the web-server.
Because I have covered a huge number of categories over the years and many pages will be included in multiple indexes/overviews/tag-clouds, I do not want multiple index pages with copies of the same text.