(cherry picked from commit 9c064dd7)

# Conflicts:
#	src/MNH/advection_metsv.f90
#	src/MNH/advection_uvw.f90
#	src/MNH/ice4_fast_rg.f90
#	src/MNH/turb.f90
#	src/ZSOLVER/advection_uvw_cen.f90
#	src/ZSOLVER/dyn_sources.f90
#	src/ZSOLVER/gdiv.f90
#	src/ZSOLVER/get_halo.f90
#	src/ZSOLVER/pressurez.f90
#	src/ZSOLVER/rain_ice_red.f90
#	src/ZSOLVER/turb.f90

Juan 19/06/2023:rain_ice_sedimentation_split.f90, Again SegFault at execition with CRAY ftn -> deactivate mnh_expand_array for CRAY ftn !!!

Juan 19/06/2023:rain_ice_sedimentation_split.f90, For pref PB with nvhpc , reintroduce mnh_expand_array & MNH_MEM_GET(ZRTMIN...) + replace data create -> present

Juan 16/06/2023:ZSOLVER/* , For Nvhpc , remove duplicate "!$acc loop" introduced by use of mnh_do_conncurrent

Juan 16/06/2023:ZSOLVER/pressurez.f90 , For CRAY ftn ,transform  mnh_do_concurrent -> DO CONCURRENNT to avoid serialisation of BR_ call

Juan 16/06/2023:MNH/* , For Nvhpc , remove duplicate "!$acc loop" introduced by use of mnh_do_conncurrent

Juan 16/06/2023:rain_ice_sedimentation_split.f90, Strange Cray SegFault -> replace mnh_do_concurrent -> array syntax

Juan 15/06/2023:ZSOLVER/turb.f90, For gfortran+openacc doesnt allo EXIT on acc loop -> replace with GZD logical test

Juan 15/06/2023:ZSOLVER/rain_ice_red.f90, For gfortran+openacc compiler SEGFAULT -> split in small acc kernels

Juan 15/06/2023:ZSOLVER/get_halo.f90, For gfortran+openacc, allocatable array not allowed in Gpu direct/host_data use_device() => TEMPORARY deactivate #define MNH_GPUDIRECT

Juan 15/06/2023:ZSOLVER/gdiv.f90,pressurez.f90, For gfortran+openacc ,replace DO CONCURRENT -> mnh_do_concurrent

Juan 13/06/2023:ice4_sedimentation_split.f90 , For gfortran+openacc ,replace DO CONCURRENT -> mnh_do_concurrent

Juan 13/02/2023: MNH modif for compile with sourceryg/2022.09-7-orig+OPENACC + optimization for CRAY ftn

Philippe 22/05/2023: FFT: deduplicate FAST_TRANSPOSE, FAST_SUBSTITUTION_2D, FAST_SUBSTITUTION_3D and FAST_SPREAD

Philippe 22/05/2023: FFT: FFT991: pass argument PWORK as a 3D array (no need to use an intermediary 1D pointer at this level)