Single-precision support is broken

[KGHustad]

The support for using single-precision floating-point numbers is not working correctly in C, C++, CUDA, and OpenCL.

Single-precision variants of mathematical functions are not used

In C/C++, exp, log, sqrt are used instead of expf, logf, sqrtf.

Double-precision literals can sneak into the code, causing implicit casts to double-precision.

Double-precision floating-point literals can sneak into the code from modelparameters.

This issue is probably hard to fix without a major rewrite of gotran or applying some very ugly fixes to the final code.