Respect enviromental CUDA_NVCC_FLAGS.
Signed-off-by: NJustin Lecher <jlec@gentoo.org>

removed __host__ from the GPU result sets, as xerion reported it to break the windows build. And fixed the cuda cmake build, as the -arch sm_13 was there twice, causing a compiler error

This commit contains part of the GSoC 2011 work of Nick Vanbaelen
who added multithreaded search support in FLANN using Intel's TBB.

Added single kd-tree index that works better for low dimensionality
points, such as 3D point clouds.

Refactored entire library so that the indexes are templated on the
distance.

Added unit tests based on gtest and pyunit to teh build system.