cuda: Fix self assignment in cudaCheckKernel macro