Any triangulation is partitionned to give parts that are disk-like, for harmonic map purposes.

There is still a bug with the mesher (the 1st GVertex (v0) of a new GEdge does not correspond to the orientation of its MLines, which implies that the 2D mesher fails to recover an edge, since it does not exist).
=> todo @JFR (a priori): fixing that
Anyway, the topology from the partitioning seems to be correct; if not, contact PAB.