Should we pin other CUDA components tightly?

During the review in staged-recipes, there was [a discussion](https://github.com/conda-forge/staged-recipes/pull/21924#discussion_r1155112007) on whether this pinning should/can be relaxed, given that we already pin `cuda-version` tightly:
https://github.com/conda-forge/libcusolver-feedstock/blob/b6c607c111cdc529a53e8465c86693aaf929bcf7/recipe/meta.yaml#L61
Opening an issue here to track the discussion.