Quentin 24/01/2019: minor correction on XCEDIS : cancellation of previous modification of XCEDIS with respect to the mixing length

Quentin 23/01/2019: modification of initialization of PM09 convection scheme at hectometric scale.The mass-flux at the surface is reduced and converge to EDKF value when resolution is meso-scale. Flag to turn on : LGZ = TRUE by default. Modifs by Rachel Honnert

Philippe 22/01/2019: replace double precision declarations by real(kind(0.0d0)) (to allow compilation by NAG compiler)

Gaelle 22/01/2019: add ARPEGE and AROME GRIB 2 (IMODEL=6 and IMODEL=7 in read_all_data_grib_case.F90) and adaptation to EPYGRAM tool

Philippe 21/01/2019: add LIO_ALLOW_NO_BACKUP and LIO_NO_WRITE to modd_io_ll to allow to disable writes (for bench purposes)

Quentin 17/01/2019: cleaning : remove of cyclic conditions-like on DP and ZA variables (useless here)

+ move IOFREEFLU and IONEWFLU to mode_io_file_lfi.f90
+ move management of NNCID and NLFIFLU to the nc4 and lfi subroutines

Quentin 10/01/2019: add a local filtering for very high slopes in PGD step (for a better chance of pressure solver convergence). In namelist NAM_ZSFILTER : possibility to activate or not (LHSLOP), choose the number of filtering iterations (NLOCZSFILTER) and the slope value (XHSLOP= Dz/Dx and Dz/Dy) where the filter is applied

Quentin 08/01/2019: correction of the TKE dissipation constant with RM17 mixing length and add RM17 in shallow convection scheme
- ini_cturb.f90 + read_exsegn.f90 : XCED constant is replaced by XCEDIS which depends on BL89 or RM17 mixing length (see Rodier et al. 2017)
- shallow_mf.f90 + all compute_*.f90 : support of RM17 mixing length in shallow convection
- bl89.f90 : add comment

Philippe 13/12/2018: split mode_netcdf module into multiple modules + add new subroutines to check if a file has been closed correctly
This is done:
*to reduce circular dependencies problems
*to have cleaner code
*to compile faster
*to detect problems with not correctly closed netCDF files

Juan 10/12/2018 : // Correction , mpi_bcast CG & CG_POS parameter & initialize INBLIGHT on all proc for filling/saving AREA* arrays

Benoit Vie 15/11/2018: bug: SMAX variable output was not written on the entire domain (it was written only in the proc domain) when multiproc used

(cherry picked from commit 0cac8488)

(cherry picked from commit bd503275)

Quentin 30/10/2018 :lima_adjust.f90, activate MEAN_CF diag in LIMA if no radiation scheme used (otherwise, in LES diag output, MEAN_CF___PROC1 = 0 even if clouds are presents)

Quentin 01/10/2018 : set_forcing.f90 , bug for NKWH>2 => move allocate(ZWHITE) before do loop on vertical levels