In a refactoring of the DELAY_LOAD stuff of Windows it seems the
Mac version was broken. On Mac there is a link to libvideostitch,
which is just a symlink to either _cuda or _opencl versions.

If the the symlink is currently pointing at the wrong one (i.e.
not the one matching our GPU), the symlink is changed and the app
is quit, user asked to launch again.

There seems to have been a refactoring done on Windows which de-
activated DELAY_LOAD completely on Mac, which made some of the
symlink changing code not compile.

Adding a new C Macro for Mac symlinks and allowing for USE_DELAY_LOAD
on Mac as well, in CMake, to restore previous behavior.

voronoi merger inherits mask merger, make destructor virtual

Fix GLEW

Create sponsoring button

Update build doc for Mac

Linux pkg sys

unused lambda capture

Now we can only use mac ports or homebrew

LINUX_PKG is now the only option

use a docker image for CI

we need this to make later versions of CUDA compile