We need to do a little finangling to get this working nicely.
For unicode, we need to switch to LuaLaTeX (I tried XeLaTeX but it
didn't seem happy for some reason), and then adjust microtype to remove
the pdfLaTeX-only flags.
Because the html and pdf exports start both at the same time,
they both experience a cache miss with the xkcd sqlite db.
Poth processes download the image, and attempt to write to the db
... with the same primary key. Hence, errors.
So, when the xkcd cache isn't populated, we wait for one process to
finish before starting the other