dg : do not initialize dofContainer to 0 by default
dg : all testcases ported to python3
gmsh : use dcopy/zcopy in fullMatrix::setAll
gmsh : add a flag to prevent initialization of fullMatrix

 -- Using PView
 -- Removing Writer for FEMSolution and ElementSolution (layer for PView)
 -- Adapting System
 -- Interpolator simplified (left until PView can handel arbitrary basis)

** Mesh: new method
 -- Can access all the coordinates of the vertices
 -- Used by Interpolator

** Simulations: Huge Cleaning
 -- Intializing by SmallFem for all simulations
 -- Merging all Projections into on code
 -- Use of FEMSolutions and ElementSolutions
 -- Rewriting ShowFunctionSpace to use only FEMSolution
 -- Convergence Tests done again (see further)

** Assembler: Removing SystemShowFunctionSpace
 -- Done in Simulation

** FunctionSpace:
 -- Changing interface in BasisLagrange
 -- Cyclic permutations reactivated in ReferenceSpace (see further)

** WARNING: Without cyclic permutation TetEdgeBasis does NOT converge !!
 -- Works in 2D
 -- Works with TetNodalBasis
 -- Fails with TetEdgeBasis
 -- ShowFunctionSpace shows weird thing for some Tets
 -- Implementation or math error ????

** Convergence Tests WITH cyclic permutation
 -- Works everywhere (line, tri, quad, tet)

boundary layers in 3D are advancing (filtering intersecting layers using rtrees)
randomness of 2D meshes is now close to be ok (we still have problems with OCC ...)

No commit message

improve reproductibility of 2D mesh algorithm (compare vertices and triangles by number, not by pointer)

when reparamOnFace fails, try to find the closest point before resorting to simply using the linear interpolation

Lagrange Basis uses ReferenceSpace -- WARNING Orientation still misses + REworked line basis + ReferenceSpace takes only connectivity into account: seems enough

Trying to delete the subclients in the event loop, to prevent leaving sockets open.

The old mechanism would lead to resource starvation when creating many subclients (e.g. in a loop).

Please report any issues with subclients -- this is quite a big change :-)

fullMatrix::gemm