Force segmentation using a fast and smart voxel based approach

[BartBruininks]

The current force segmentation show good results, however, it is implemented in a rather inefficient manner. The recursive nature is nice, but slow and new distance calculations are performed over and over again. It would be better to start with making a mask for all that is non-segmented and then perform the neighbor search and do the largest distance calculations first. Then we can subsample and update the segment IDs in an iterative manner. This would prevent recalculating the distances over and over again and would focus on updating segment ids and our current most dominant segment in the surrounding. Subsampling the max distance matrix with a shorter cutoff (already sorted in distance, so just array indexing). Something like that.