Fix MPI memory leak

See merge request !192

For reproducing sums, the allocated MPI operation was not freed.

Add the new gnuplot raw writer

See merge request !191

It works in any dimension, coordinates are not saved. It's main
purpose is to produce scatter plots.

Also add a few documentation.

It handles IBoundaryDescriptor data and its access functions

Add tensorProduct for DiscreteFunctionP0

See merge request !190

The tensor product is performed cell by cell

Also give access to the tensor product in the language. In this
special case vectors have to be of the same dimension since non-square
matrices are not allowed there

Also add related tests

Fix command line option for reproducible summations

See merge request !189

Actually, the "naive" (non reproducible mode) could not be used. The
code was always computing reproducible sums, which is the default.

Add new CLI option to allow parallel output

See merge request !188

Only write faces and faceoffsets fields in pvtu when needed

See merge request !187

Actually, it seems that reading of general polyhedral cells is broken
in early versions of Paraview
- is it a change in the format (poorly documented...)
- or just a Paraview's bug?

Fills the vtkGhostType field to fix parallel rendering

See merge request !186

Fix max of array types in the case of all non-positive values

See merge request !185

Replace instances of -std::numeric_limits<T>::max()
by std::numeric_limits<T>::lowest() which is more generic

Thanks to Philippe for noticing it.

This was a classic misused of std::numeric_limits<T>::min()...

Fix tag for signed integers in vtu outputs

See merge request !184

All integer values were marked as unsigned.
Was not noticed since output data were probably too small (<2^31)

Release v0.5.0

See merge request !183

Its been a while since 0.4.1.

Lots of changes, here we mainly change release tag to mark the
important change in mesh handling (using variant instead of interface)

gcc and clang-16 do not agree with syntax in case of friend concept
template functions...

Get rid of IMesh (mesh interface class) and use MeshVariant instead

See merge request !182

It now uses mesh type template argument instead of space dimension

-add MeshConcept to simplify compilation error
-add mesh_type_t template utility to simplify
-simplify code (dealing with constness of mesh types was not that easy)
-check constness (at compile time) to help code writing in boundary utilities

On the way:
- add bunch of helpers to MeshVariant
- add MeshConcept
- add is_mesh_v
- rename is_polygonal_mesh -> is_polygonal_mesh_v

Mesh<Connectivity<Dimension>> now writes simply Mesh<Dimension>

Also remove dimension template parameters for Mesh*Interface classes
forgot previously

- remove template parameters that were useless (mostly when node
coordinates are useless)
- replace dimension template parameter by mesh type (prepares the use
of new kinds of meshes)