Here I'm just testing yet another algorithm to computing edges from polygons.
Compared to the existing one it makes use of the assumption that most vertices have 4 or less incident edges.
Furthermore, it does not deal with the case of keeping existing edges and/or flags around.
It also does not do validation of existing edges.

It's supposed to be optimized for the case when geometry is created procedurally but it is difficult
to compute the edges directly so that they have to be derived from the polygons.

In my tests the new approach is 3-4x faster for sufficiently large meshes (in the single and multi-threaded case).