[CI] Publishing pipeline fixes continued

[khalatepradnya]

This PR follows-up the work from #3544

Successful pipeline - https://github.com/NVIDIA/cuda-quantum/actions/runs/19126706667

• Updated workflow and validation scripts to use full CUDA version strings (e.g., 12.4.0, 13.0.2) for conda channel selection for compatibility with available CUDA packages
• Improved detection logic for CUDA versions in setup.py by including CUDA 13 library names, ensuring correct identification of the environment. (python/metapackages/setup.py)
• Refactored CUDA library path setup in cudaq/__init__.py to dynamically construct and set the CUDAQ_DYNLIBS environment variable based on the detected CUDA version, supporting both CUDA 12 and CUDA 13 and handling missing libraries gracefully. (python/cudaq/__init__.py)
• Updated Dockerfile installation logic to install the correct set of CUDA Python packages depending on the major CUDA version, improving container compatibility and reducing installation errors. (docker/release/cudaq.ext.Dockerfile)
• Enhanced notebook validation workflow to create a dedicated Jupyter kernel and virtual environment for container-based notebook execution, allowing notebooks to install their own dependencies and improving isolation and reproducibility. (scripts/validate_container.sh, docs/notebook_validation.py)
• Added missing Python packages (python3-venv, python3-pip) to the publishing workflow setup to ensure required tools are available for builds and validation. (.github/workflows/publishing.yml)
• Metapackage validation: (1) explicitly specify the conda channel, don't assume the .0 patch version in the channel name (conda 13.0.0 channel is not a real CUDA 13 channel); (2) install torch (during validation) matching the version of cuda-q wheel dependencies (prevent conflicting dependencies).

[github-actions]

CUDA Quantum Docs Bot: A preview of the documentation can be found here.

[sacpis]

Overall this PR looks good. Just have one comment. Thanks @khalatepradnya.

[github-actions]

CUDA Quantum Docs Bot: A preview of the documentation can be found here.

[bmhowe23]

Wow, I assume we need to make a similar change to CUDA-QX? Will this need to go int the docs?

[1tnguyen]

Definitely. I'll update the doc in a follow-up PR. Currently, it has a similar note about using the torch nightly channel for CUDA 12.8. I'll check whether it's still needed and add a note for CUDA 13.

[bmhowe23]

while cuquantum-cu13 depends on cuBlas 13.1.

I do not see this dependency. When I install in a fresh environment, the pipdeptree that I see is this:

cuquantum-cu13==25.9.0
├── custatevec-cu13 [required: ==1.10.0, installed: 1.10.0]
├── cutensornet-cu13 [required: ==2.9.0, installed: 2.9.0]
│   └── cutensor-cu13 [required: >=2.3.1,<3, installed: 2.3.1]
└── cudensitymat-cu13 [required: ==0.3.0, installed: 0.3.0]
    ├── cutensor-cu13 [required: >=2.3.1,<3, installed: 2.3.1]
    └── cutensornet-cu13 [required: ~=2.9, installed: 2.9.0]
        └── cutensor-cu13 [required: >=2.3.1,<3, installed: 2.3.1]

Where do you see that cuquantum-cu13 requires cuBlas 13.1?

And the prerequisites that I see on https://docs.nvidia.com/cuda/cuquantum/latest/custatevec/index.html simply say 12.x and 13.x for CUDA 12 and 13, respectively.

[bmhowe23]

I do, on the other hand, see where we/CUDA-Q are declaring a dependency on 13.1, but perhaps we shouldn't be doing that?

cuda-quantum/pyproject.toml.cu13

Line 25 in 9336ee2

'nvidia-cublas ~= 13.1',

[bmhowe23]

I think #3589 needs to be incorporated to address the Torch issue. That allows us to use a released version of Torch rather than forcing our users to use a nightly version of Torch.

[bmhowe23]

LGTM, but it wouldn't hurt my feelings if @bettinaheim reviewed, too.

[khalatepradnya]

Merging to facilitate package creation...

[github-actions]

CUDA Quantum Docs Bot: A preview of the documentation can be found here.