Philippe 13/02/2018: IO_GUESS_DIMIDS_NC4: manage variables with CLBTYPE/=NONE + add a return value + problems are now WARNING instead of FATAL

* write NF90_INT1 instead of NF90_INT for logicals
* use IO_WRITE_FIELD (not yet done everywhere)
* if NTYPE is unknown => fatal error
* do not try to read JPVEXT (never written)

Philippe 09/03/2018: lfi2cdf: remove redundant fields of workfield (present also in TFIELDDATA) + remove read_data_lfi

Philippe 23/02/2018: lfi2cdf: add CPROGRAM + some initialisations to later use file and fields structures from MNH

(necessary for LFI2CDF)

Philippe 22/02/2018: lfi2cdf: works with more situations (not yet perfect) + nc4 files have a .nc suffix

Philippe 20/02/2018: SURFEX: add special treatment if KP is zero size (to prevent problems with XLF)

Philippe 16/02/2018: SURFEX: bug correction: allocate some work arrays to 0,1,1 instead of 0,0,1 (crash with XLF)

Philippe 13/02/2018: re-added #ifndef MNH_SP4 to prevent problems when compiling on Blue Gene/Q with -qautodbl=4

* always set MNH_REAL (8 by default)
* MNH_REAL has to be set to 4 or 8 (and not R4 or R8)
* XYZ is slightly modified to contain ...-ARCH-RxIy-...
* pass MNH_REAL to pre-processing
* replace MNH_REALS by MNH_REAL

* corrected intent in READ_ALL_DATA_GRIB_CASE
* workaround for compiler bug (Intel 2013.1) in IO_WRITE_HEADER
* use own NetCDF (bug in default version of Ada (4.1.3))

Philippe 09/03/2018: IO/LFI: add NFILENAMELGTMAXLFI parameter and use it to force size in LFI file of MY_NAME and DAD_NAME
This is done for backward compatibility.