So I spent some of the Victoria Day weekend on PyOpenGL. Biggest changes are:

* got cibuildwheel running on github actions so that we can get a matrix of (manylinux) wheels built for accelerate
* automated upload from github actions to pypi
* investigate the egl regressions; so far all of them seem to be related to something changing in PyGame. I constructed a raw-xlib test-case runner that shows the actual EGL bits work, but it seems likely that something in SDL or Pygame is creating an EGL surface internally
* merged a bunch of pending pull requests

The cibuildwheel process takes 2+ hours, but having the entire pipeline automated should make it easier to release new PyOpenGL versions.

One issue I encountered that I likely need to look into is that until I updated numpy on my test environments I saw object-header-mismatch failures on the manylinux accelerate builds. My assumption there is that it's a numpy version issue, so we may need to look at embedding the numpy version in the accelerate dependencies.