* RFC 4122 does not guarantee octet 0 to be non-zero.

* This partially reverts "added support for C89/90 preprocessor" which only
  covered up an already bad situation.

Using the C preprocessor on Fortran sources does ultimately not work
since it doesn't know about Fortran-specific features like the _OPENMP
macro and is chronically likely to pick up the cpp belonging to
another compiler (e.g. to system gcc when compiling with xlf on Linux/ppc).

* reshGetVal never returns NULL.

* This way CDI can still be built with pgcc 11.x.

* Also extend Fortran 2003 interface generator accordingly.