rename {p,{int,ext}}_comm to {Int,Ext}, fixing relevant trelo card

See merge request dobrzyns/ParMmg!8

Bug fix in -hsiz case : in mmg3d, the hmin/hmax are not scaled in scalem function (because it is done after in the soltruncature) but in parmmg, we need to scale the hmin at each iteration of the parallel mesh adaptation loop (libparmmg1)

replace PMMG_SUCCESS/PMMG_FAILURE return codes with 0/1

Add a preprocessor variable to define the supposed maximum number of groups adjacent to a given group.

Change of the return values in grpsplit_pmmg.c file to be consistant with the other functions. 0 : failure ; 1 : success ; -1 : strong failure
Error handling in libparmmg1.c

Do not retrun immediately in grpSplit if we don't have any group to split : got to the end of the function and check the communicator consistency (it allows to check it without any problem when 1 proc split groups into sub groups and the other one not.

Patch error in check_intNodeComm : the coordinates were compared with != operator instead of a comparison with a given epsilon.

Remove an MPI deadlock added be the call of the PMMG_check_extNodeComm/FaceComm functions in grpsplit: if 1 proc needs only 1 group and the other needs multiple groups, the function is called only by one proccess...

Change the way we scale the mesh in parmmg to be more consistant (needed for the node communicator construction from the faces ones) : now we scale the mesh for the analysis step (parmmg.c) then we unscale it. And we scale it again for the mmg call. It is immediatly unscaled.

Patch missing header inclusion and add reduce to avoid deadlock in the build of the node communicators from the faces ones.

Feature/build node comm

See merge request dobrzyns/ParMmg!2

Add the free of the mesh without triangles if we don't have enough memory to rebuild them