add missing device annotation for transpose

[tukss]

When compiling for CUDA using clang++ (version 22.1.5), I got this error:

[...]/include/codi/expressions/../tapes/interfaces/../../traits/computationTraits.hpp:80:39: error: 
      reference to __host__ function 'transpose' in __host__ __device__ function
   80 |       return TransposeImpl<Jacobian>::transpose(jacobian);

nvcc compiled the same code with warnings about the missing annotation but consistently produced NaNs in the gradient output.

Adding CODI_INLINE on the transpose fixes both problems for me and I get correct gradients in a CUDA test case using the codi::RealForwardCUDA type.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[MaxSagebaum]

Thanks for the pull request. Since it is only a minor change, I moved forward and merged it.