Anyway, thanks for the links and the thoughts! Amusingly almost all of the links were already visited! I've looked several times at your library for geometric constraint solving in Python and almost thought about using it for my own projects. In the end however I think SolveSpace is the best choice -- it is solid, actively developed and fast.
Your ambition of having the drawings be singularly specified by their constraints is absolutely admirable, but from my experience it is simple not feasible nor desirable. First, the optimization problem is difficult to solve and it will be difficult to specify it such that there is only one unique solution. Secondly, I think the human provided prototype (through a GUI) is a critical part of the design process which makes this technology usable. But I understand the desire to have drawings specified only by constraints.
In the long term, if I am allowed to dream, I'd like to see the development of a more general purpose constraint-based and parametric drawing library. Right now I am focused on PCB footprints, but many of same ideas are useful in electrical diagrams, optical diagrams, GUIs, page layout etc etc. We'll see how far I get ;)