v0.7.2

While working on a technique to speed up nearest neighbor queries, it became clear that the benchmarks were not representative for showing what the performance would be when matching between two different point clouds. Instead, they showed the performance of queries when the same point cloud was used for both building a tree and querying it. A benchmark using two different clouds is more indicative of tree performance and PicoTree was not the fastest in this scenario. This release addresses the issues with both the benchmarks and query performance.

Changes:

• Improved query speed for both the C++ and Python libraries at the price of reduced metric support.
• Removed L2 and EigenL2 metric support. The KdTree now only supports distance functions that don't apply a final exponent when determining the length of a vector (e.g. ^(1/2) for the L2 metric).
• All benchmarks now test matching performance using two different point clouds.