Corrected bug in obs_write.F when computing final misfit

[arango]

• A bug was corrected in obs_write.F when computing the final misfit in I4D-Var. The TLmodVal is included and we need only to have:

          IF (wrtMisfit(ng)) THEN
            DO iobs=Mstr,Mend
  #  if defined IS4DVAR
              misfit(iobs)=ObsScale(iobs)*SQRT(ObsErr(iobs))*             &
       &                   (NLmodVal(iobs)-ObsVal(iobs))
  

  This is just a diagnostic quantity for output and doesn't affect the I4D-Var computations. Many thanks to Ha Joon Song for nicely ​reporting this bug.

• Updated the makefile to add version 3.82 of gmake to the NEED_VERSION macro. The gmake 3.82 was released earlier in the summer.

• Updated the documentation in build.sh and build.bash. Notice that it is possible to compile with a serial NetCDF-4/HDF5 library in distributed-memory (MPI). The parallel version of these libraries is only required for parallel I/O when the C-preprocessing option PARALLEL_IO is activated in ROMS. Many thanks to David Robertson for bringing this to my attention.

• Updated the UPWELLING test case so we can use either EW_PERIODIC or NS_PERIODIC. We just needed to reformulate the bathymetry (ana_grid.h) and wind stress (ana_smflux.h) forcing to the opposite direction in NS_PERIODIC applications. Still the EW_PERIODIC is the default configuration.

• Updated documentation in distribute.F about why we cannot USE mod_netcdf in mp_ncread and mp_ncwrite because cyclic dependencies.