in the long run, at better data structures is needed for detecting edges.
The collision tree edge flag edges and provide the information but it is still doing a one triangle at a time contact calculation and then trying to remove the redundant contacts.
what is needed is a better data structure that encode convex edge strips so that the information
provided contain vertex, triangle indices an edge connectivity.
this way the contact at an edge will be calculated from the strip at once instead of one contact per
face and the collapsing the resulting contacts.
Maybe for some newer version 3.20 or better for now this is OK but it bothers me a lot.