separated from Octopus specific code and can be reused for other
purposes. This makes the Octopus part cleaner.

 Looks like g95 and pgf do not like public/private statements for specific derived data types members.

 * It is now possible to use the same partition as the one of PFFT when running parallel in domains.
   This implied several changes in the order by which things are initialized: the cube used in the poisson solver has to be initialized before doing the partition of the mesh.

 * I took the opportunity to rewrite part of the PFFT stuff and moved most of the things stored in pfft_t, to the cube or to the cube functions. The medium-term goal of this is to hide as much as possible the use of PFFT.

* Removed N. Suberviola from the copyright. With my last changes, all
his code has been removed.

the fourier_spectrum routine, and so it means that the compressed
sensing method can be used for this too.

* Added a missing error code.

* Renamed some constants.

* Reimplemented f90_cl_has_extension in Fortran.

OpenCL independent of Octopus specific code.