Don't build with MKL by default in setup.py

• Don't build with MKL unless --use-mkl' flag is passed or USE_MKL` environment var is set
  • MKL is hard to install if you're not using conda and I don't want to break pip users. I'll enable it by default in conda.

Bump version of numpyeigen

• Bump numpyeigen version which now inclcudes Python.h before numpyarrayobject.h.
  • This should fix older versions of numpy

Make numpyeigen a CMake external project

• git submodules suck, move numpyeigen to be a CMake external project instead.
• Make MKL required if it is enabled

Intel MKL Support

• Added optional MKL support from CMake
• Cleaned up setup.py so things should be buildable from conda

Initial release of pcu

First set of functionality in pcu:

• Various ways to generate point samples on meshes
  • uniform random
  • Poisson disc sampling
  • Lloyd relaxation
• Hausdorff distances and pairwise nearest neighbor queries