Use -fvisibility=hidden.

This allows gcc to optimize function calls to globally visible, but not exported
(using PMOD_EXPORT) functions to a direct call. Saves a few % CPU, and also decreases
 the size of the binary.

However, it is now important to use PMOD_EXPORT correctly on any system, not only Windows.

It also significantly speeds up loading of dynamic modules, but that does not generally speaking
use any significant amount of CPU anyway.