ci: source CUDA ver. from nvcc

[Borda]

What does this PR do?

Use nvcc instead of torch to source base CUDA docker images. See the reason why not to source nvidia-smi: https://github.com/NVIDIA/nvidia-container-toolkit/issues/300 BUT nvcc is missing in runtime docker images...

Before submitting

• [x] Was this discussed/agreed via a GitHub issue? (no need for typos and docs improvements)
• [ ] Did you create/update your configuration file? (in case you are adding new integration)
• Did you set runtimes in config for GitHub action integration?
• [ ] Are all integration tests passing?

[codecov[bot]]

Codecov Report

Merging #106 (5e0e540) into main (d30ef2f) will not change coverage. The diff coverage is n/a.

Additional details and impacted files

@@         Coverage Diff         @@
##           main   NVIDIA/nvidia-docker#106   +/-   ##
===================================
  Coverage    80%    80%           
===================================
  Files         2      2           
  Lines       252    252           
===================================
  Hits        201    201           
  Misses       51     51