Upon investigation the issue was installing Theano and having it ‘play nice’ with PyMC3 on Windows 10, rather than problems with the model itself.
Depending on whether I tried installing everything via conda, conda-forge (e.g. theano, m2w64-toolchain, mkl-service, etc), or manually installing MinGW for the g++ compiler (as suggested here), I variously encountered errors very similar to those described here and here. Typically the model would stall at 3.40% or so, regardless of the number of cores, chains, priors, inits.
My current solution is not ideal but working well, sampling within 15 seconds and yielding results similar to @nkaimcaudle (thanks btw - that looks correct as per the text!). I followed these instructions, setting up a virtual environment using somewhat older versions of Python and PyMC3 (both 3.5). EDIT: Actually, even this is only working with one core. Trying to add more throws this error 
.
I’ll keep trying to fix the issue in a separate virtual env (just came across another post which may be helpful) and will report back if I find any solutions, but setting things up on Windows 10 unfortunately remains a bit of a headache for now. Thank you thank you @junpenglao, and super excited for PyMC4!