Linking libcuda.so into environment

[DoeringChristian]

Hi there,

I hope this is the right place to ask about this.
I've been using pixi.sh for my python projects for a while now. I'm often using tinycudann, and pixi finally allowed me to get a somewhat reproducible environment goint, with which I can install it. The only issue is that, when compiling tinycudann (and other cuda related projects), they require the libcuda.so shared library (this is the driver library as far as I can tell) to be present in the LIBRARY_PATH. Unfortunately, neither the cuda system-requirements package nor any other cuda package link this library into the conda environment. Atm. I have to resort to adding my whole /lib64 folder (on fedora) or the /usr/lib/x86_64-linux-gnu path on ubuntu to my LIBRARY_PATH, which might expose other libraries as well, that I don't want any programs to be built against.
Therefore, I wanted to ask if there is a better/more robust way to handle this?
Would it for example be possible for the cuda dependency to link libcuda.so into the environment, or should I ask somewhere upstream?

Thanks for your Help!

Kind regards
Christian