I think we can forget about dt nurbs, for the reasons you explained above and also because it would be simpler to write everything from scratch than trying to read/update it or porting it to Fortran.
SISL is excellent but it misses for example the curve rotation feature.
For my needs it would be sufficient definition, fitting, analysis and interrogation of b splines and bezier curves/surfaces. They have to work on simple cases, full CAD capabilities are not required when doing simple FEM /CFD analysis.
Nurbs are really not needed, lots of people are happy not having an exact representation of a cylinder.
What is your time estimate on writing such a library by 1/2 people?