ncjoin with simple precision
ncjoin is not working with single precision, so we force output in double in cppdefS_dev.h since v1.3 The previous modification (conversion of real8 to real4 when calling assembly_real) is actually not working, and generate empty files. This modification was done bacause of the argument mismatch when calling assembly_real, which turned to a compilation error with gfortran >= 10. The modification is not needed if we use the compilation option -std=legacy, so we can revert it. Note that relying on a compilation option is questionable.