Improve the post-processing performances

[delhomer]

For now, the post-processing (when required) is run in a "brute-force" mode. We could certainly be smarter, for example by parallelizing the treatment (as we do for feature extraction).

Some benchmarking may be done from the PCL library: http://pointclouds.org/documentation/tutorials/resampling.php#moving-least-squares.