integration points (order 8 meshes 10 times faster) + correct a bug in
dg for non iso-parametric in implicit

No commit message

(fracture in bending OK) 3) lua version of code 4) a test case with 3rd
order quadrilateral element. 
Some modifications are made in linearSystem (add a function to compute
norm of RightHandSide) and in dofManager (add a function to have the
fixed dof outside dofManager)

physical tags, partitioning, second order prisms

No commit message

fix multiple cut

meshPartitionOptions::partitionByExtrusion assigns all extruded elements to the same partition as the source element

* issue error for zero volume elements in FEA export
* reset element number between 2D/3D in FEA export

No commit message

free to improve...)

pretty please do not just "throw;" when you don't know what to do

+ constify MElement some more

single file IO)

TWO BIG WARNINGS:

1) The MSH file version has been bumped from 2.1 to 2.2 because if a
mesh has partitions, the meaning of the tags has changed. Tag 3 used
to be the partition id; now tag 3 contains the number of partitions
(self+ghosts, if necessary), and the next tag(s) contain(s) the
partition(s) id(s). Ghost partitions have < 0 ids. The (new) default
behaviour is to not save partition numbers by default if there are
none.

2) There is an issue for split-file IO (one file per partition), due
to our on-the-fly renumbering of the elements (we would need a global
pass+store an inverse multimap to have the correct ghost element
numbers in a split-file scheme). The problem comes from the multiple
physicals -> multiple copies of the same MElement* in the MSH file +
our refusal to store an index in MElement. We need to discuss this
before we implement split-file IO in parallel...



PS -- I also changed the way we write binary files. We now save blobs
of single elements. This is fully compatible with old versions, but
might slow down reading large files. I did not see any performance hit
on my machine, but this should be tested more thoroughly. Third party
codes can still write larger binary blobs if they want to, but Gmsh
itself now writes small blobs, which makes the writeMSH routine much
simpler.

- renamed functionSpace into polynomialBasis (functionSpace will be much higher level)

renamed most gmsh* classes