optimize CFLAGS from better performance

[breznak]

I've noticed this library is built without any performance related compiler flags. Or at least it is not documented anywhere. It would be interesting to investigate performance gain from using eg: CFLAGS="-Ofast -march=native -fuse-linker-plugin -flto-report -pipe" LDFLAGS="-O3 -flto" python setup.py build for the source install. Note: for binary distribution remove -march=native or replace it with some more generic CPU architecture.

I would have tested this, but I'm stuck at #38