1D and 2D moving mesh algorithm

I have added a new hydro scheme, `SHADOWFAX_SPH`, that uses a finite volume method on a moving mesh. The scheme is very similar to `GIZMO_SPH`, except that during the density loop an actual Voronoi cell is constructed for every particle, which is then used to get a volume to update the primitive quantities, and faces through which fluxes are exchanged. There are also some small changes in the way gradients are computed, and in the way the velocity of the particles is set.

The Voronoi cell construction is completely different for 1D, 2D, and 3D, and currently only 1D and 2D have a stable implementation. 3D should compile, but does not produce reliable results (yet). However, I think it would be good to merge this in already, to prevent it from drifting too far apart from `master`.

The current implementation is hydro only, but it should be very straightforward to add gravity, similarly to the way it is done in `GIZMO_SPH`.

Below the results for the 2D Sod shock and Sedov blast test (the noise on the density is caused by the fact that the initial conditions contain mass and not density). The 2D Noh test crashes because of time stepping issues; I think we really need a time step limiter to make it work properly.

![SodShock](/uploads/641796ae26cf045be5ce824bb698b77a/SodShock.png)

![Sedov](/uploads/96be4bbeb7635c329b78f89e0e79239c/Sedov.png)

See merge request !321

Added some more comments to 3D algorithm, and fixed a bug that affected some of the degenerate cases.

Autotools update

Attempt to handle the case when HDF5 is available from the environment (CPPFLAGS, LIBS etc.)
but no h5cc or h5pcc commands are available.

See merge request !322

Replaced global variables by a new hydro_space struct that is part of space. This struct is then passed on to hydro_init_part, and used to initialize the Voronoi cells. For non SHADOWFAX_SPH schemes, this struct is empty. Replaced random_uniform declarations in some unit tests with the tools.h version. Added some more comments to 3D Voronoi code; it is now almost completely documented (meaning I almost completely understand it now).

Only repartition when required

Only repartition when the previous step processed some large fraction of all the
particles, and then only when the loads between the ranks are out of balance.
This is for several reasons:

  * Repartitioning is expensive, so should only be done when necessary.
  * Frequent repartitioning with multi-dt is not necessary (for the EAGLE volumes anyway).
  * It is more representative to check the load balance when all tasks have been ran.

The load balance is determined from the user CPU time per step (including
the CPU time from all threads). We exclude the system time as that is not
down to processing and tends to even out the ranks artificially, much as
elapsed time does (since we wait for all the MPI tasks to come together).

The load imbalance allowed is determined by the parameter `DomainDecomposition:trigger`,
this can also be set to a number greater than one, in which case the old
repartitioning scheme of every 'trigger' steps will be used (previously trigger
was always 100).


See merge request !290

Added a lot of comments to 3D Voronoi algorithm, fixed bug in testVoronoi3D. Disabled expensive checks. Some basic 3D moving mesh tests now work.

If the h5cc or h5pcc commands are not available don't assume that HDF5 cannot be used, test it instead. We lose the version string and maybe Fortran support.

Don't use the h5cc command when determing if parallel HDF5 is available, check the H5_HAVE_PARALLEL symbol instead

Set the unnecessary dimensions of the box in 1D or 2D to safe values to prevent unnecessary crashes due to lack of cells along this particular axis.