The CUDA docs are not perfect. The usual advice I offer if you find an issue or want to see an improvement is to file a bug.
You could use godbolt compiler explorer. I use it sometimes. For example, for the code I have posted above, if I put it into godbolt and select CUDA 12.0.0 (the version suggested in the blog), it seems to compile correctly. (if you switch the compiler version to 11.8, for example, the compilation will fail.)