For context/the uninitiaatd, I think of pandoc as the "postgres" of markdown-to-<blessed_format>: you can build your own text-format-converter, but oftentimes you can get 90% of the way there with pandoc, and fill in the rest with custom filters and elbow grease.
To whet your appetite, I have a personal site that checks your boxes, where the build script is basically one long invocation of pandoc and xargs. In my experience and that of others', building on top of pandoc can get you to quite satisfying places and workflows (e.g. https://www.gwern.net/About).
(Also, fwiw, we use it at my place of work for writing versioned technical documentation, as well.)
Keep on plain-texting! :)
I ended up exporting all of my old blog posts from WordPress. Rebuilding the entire site (~370 pages) takes about 4 seconds.
The biggest problem I found was that I had to modify tufte.css to make it work well on mobile.
Care to post your site?
Mine, if you are curious: https://sheep.horse/
In my case there is a certain irony in writing it in Python the lazy way since I'm a full stack dev who writes typescript, js and PHP for work.
Never trust a builder with a finished house I guess applies, I didn't want anything I'd have to maintain.
edit: oh wait that's what the Tufte css is! What tweaks did you end up making?
It took me a lot of fiddling before I managed to find an acceptable compromise.
If I was going to start again I might consider implementing the margin notes differently. They interact oddly with each other and image captions.
Has syntax highlighting as well, beyond that I wrote it myself because it let me do exactly what I wanted the way I wanted.
Code is here if it's any use to you https://kopy.io/ibFMc
It supports Jinja templates and you can put meta data in the header so you can choose what template you want, override the title etc.
If you don't override the title 10-tricks-to-foo becomes "10 tricks to foo" automatically
I've used that script to write technical documentation as well, I just swapped out the template and everything else stayed the same, in terms of time to value reward it's up there for the most useful code I've written in a long time.