For scaling up, we can move to cKDTree
from what we are currently using (KDTree), by modifying cg/kdtree.py

This breaks some tests due to different ordering of returns. Perhaps fixing the tests and getting this implemented is a good item for the SciPy sprint?