Fix floating point exceptions for systems without working feenableexcept

This merge request provides a new way to use floating point exceptions based on the idea of Dave Richards from the ecl-devel mailing list of explicitly checking for floating point exception bits whenever creating a float. This should work when feenableexcept is not declared or not working properly (i.e. on Mac OS and possibly also on Windows) or when the underlying hardware doesn't generate floating point exceptions. I also included a new check in the configure script to detect the latter case.