CP - fix to mgga with gpu

The calculation of the meta-gga term of the force in CP (CPV/src/forces.f90) was missing in the gpu double of the routine. Now it has been added. I have done it on CPU in order to avoid code duplication. In a second time, while keeping on the transition to openACC, it will be ported too with acc directives.
This is the second MR submitted because the first one for mysterious reasons had the cmake-gnu pipeline failing (the commit mod is entirely protected by #ifdef __CUDA)