Feature title:  Add facade NVIDIA Python packages to CUDA base image

Feature Overview:
Upstream PyPI packages like PyTorch depend on several NVIDIA python wheels like nvidia-cuda-runtime-cu12. In downstream we use system packages for CUDA stack. Our base images and downstream Python index do not provide the packages. We should pre-populate the CUDA base image with facade packages that point to system libraries and headers.

Product(s) associated:

RHAIIS: no
RHEL AI: no
RHOAI: yes

Goals:
Provide high-level goal statement with user context and expected user outcome(s) for this Feature

• Our deliverables look more similar to upstream content.
• Users can figure out CUDA version and packages with pip / uv list (like with upstream content).
• Users get better feedback when they try to mix upstream with downstream content. pip will refuse to remove or update a package

Requirements:

• CUDA base image comes with facade packages pre-installed.
• package name and package versions are taken from RPM
• all packages have a proper METADATA and INSTALLER file in the dist-info directory but no record file
• lib and include subdirectories point to correct system locations (nvshmem is different!)
  • nvidia/cuda_runtime/lib -> /usr/local/cuda/lib64
  • nvidia/cuda_runtime/include -> /usr/local/cuda/include

Since the packages need to match RPMs and have a symlink to external resources (not supported by wheel format), the packages could be created by a script.

Done - Acceptance Criteria:
All nvidia Python packages typically used by Torch and vLLM are present in the base image

Use Cases - i.e. User Experience & Workflow:
Users try to replace nvidia Python wheel and get a clear error message.

Out of Scope:
TBD

Documentation Considerations :
Presence of package need to be documented, at least internally in base image repo

Details

List for Torch 2.9.0

nvidia-cublas-cu12==12.8.4.1
nvidia-cuda-cupti-cu12==12.8.90
nvidia-cuda-nvrtc-cu12==12.8.93
nvidia-cuda-runtime-cu12==12.8.90
nvidia-cudnn-cu12==9.10.2.21
nvidia-cufft-cu12==11.3.3.83
nvidia-cufile-cu12==1.13.1.3
nvidia-curand-cu12==10.3.9.90
nvidia-cusolver-cu12==11.7.3.90
nvidia-cusparse-cu12==12.5.8.93
nvidia-cusparselt-cu12==0.7.1
nvidia-nccl-cu12==2.27.5
nvidia-nvjitlink-cu12==12.8.93
nvidia-nvshmem-cu12==3.3.20
nvidia-nvtx-cu12==12.8.90

Pip refusing to update a package:

$ rm nvidia_cuda_runtime_cu12-12.8.90.dist-info/RECORD
$ echo "RHAI system package" > $ rm nvidia_cuda_runtime_cu12-12.8.90.dist-info/INSTALLER
$ pip install -U nvidia_cuda_runtime_cu12
...
ERROR: Cannot uninstall nvidia-cuda-runtime-cu12 12.8.90, RECORD file not found. Hint: The package was installed by RHAI system package.