this has negative impact on surfaces count in certain scenario (big axial surface of easily mergable triangles)
it wins predecessor once we rotate such surface
so may be consider choosing sorting plane per lump or mergables
algorithm itself is dependent on test data ordering, thus best option would be tweaking it, not prerequisites
`redundant indexes suppressed` count is always worse with this due to sorting nature, sorting per axial plane helps
assuming this optimization is not important and case dependent