[BUG] Calculating ground truth on GPU in cuvs-bench fails in 26.04 docker image

Failed to compute ground truth in cuvs-bench 26.04. Same command worked previously in 26.02.

docker run --gpus all --rm           --user root           --shm-size=16GB           --entrypoint /bin/bash           --workdir /data/benchmarks           -v /raid/cuvs-bench-runner:/data/benchmarks           rapidsai/cuvs-bench:26.04-cuda13-py3.12 -c "python -m cuvs_bench.generate_groundtruth         /data/benchmarks/datasets/BeIR-trec-news-generated-queries-1.5M/base.fbin         --queries=/data/benchmarks/datasets/BeIR-trec-news-generated-queries-1.5M/query.fbin         --output /data/benchmarks/datasets/BeIR-trec-news-generated-queries-1.5M -k 10
        "
          
Traceback (most recent call last):
  File "<frozen runpy>", line 198, in _run_module_as_main
  File "<frozen runpy>", line 88, in _run_code
  File "/opt/conda/lib/python3.12/site-packages/cuvs_bench/generate_groundtruth/__main__.py", line 58, in <module>
    xp = import_with_fallback("cupy", "numpy")
         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/conda/lib/python3.12/site-packages/cuvs_bench/generate_groundtruth/__main__.py", line 44, in import_with_fallback
    module = importlib.import_module(primary_lib)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/conda/lib/python3.12/importlib/__init__.py", line 90, in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "<frozen importlib._bootstrap>", line 1387, in _gcd_import
  File "<frozen importlib._bootstrap>", line 1360, in _find_and_load
  File "<frozen importlib._bootstrap>", line 1331, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 935, in _load_unlocked
  File "<frozen importlib._bootstrap_external>", line 999, in exec_module
  File "<frozen importlib._bootstrap>", line 488, in _call_with_frames_removed
  File "/opt/conda/lib/python3.12/site-packages/cupy/__init__.py", line 18, in <module>
    from cupy import _core  # NOQA
    ^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/conda/lib/python3.12/site-packages/cupy/_core/__init__.py", line 5, in <module>
    from cupy._core import core  # NOQA
    ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "cupy/_core/core.pyx", line 1, in init cupy._core.core
  File "cupy/_core/_routines_math.pyx", line 7, in init cupy._core._routines_math
  File "cupy/_core/_reduction.pyx", line 1, in init cupy._core._reduction
  File "cupy/_core/_cub_reduction.pyx", line 260, in init cupy._core._cub_reduction
  File "/opt/conda/lib/python3.12/site-packages/cupy/_environment.py", line 86, in get_nvcc_path
    _nvcc_path = _get_nvcc_path()
                 ^^^^^^^^^^^^^^^^
  File "/opt/conda/lib/python3.12/site-packages/cupy/_environment.py", line 198, in _get_nvcc_path
    cuda_path = get_cuda_path()
                ^^^^^^^^^^^^^^^
  File "/opt/conda/lib/python3.12/site-packages/cupy/_environment.py", line 78, in get_cuda_path
    _cuda_path = _get_cuda_path()
                 ^^^^^^^^^^^^^^^^
  File "/opt/conda/lib/python3.12/site-packages/cupy/_environment.py", line 151, in _get_cuda_path
    conda_cuda_path = _get_conda_cuda_path()
                      ^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/conda/lib/python3.12/site-packages/cupy/_environment.py", line 130, in _get_conda_cuda_path
    cuda_path = os.path.join(conda_prefix, 'targets', arch)
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "<frozen posixpath>", line 76, in join
TypeError: expected str, bytes or os.PathLike object, not NoneType

Works in 26.02:

docker run --gpus all --rm           --user root           --shm-size=16GB           --entrypoint /bin/bash           --workdir /data/benchmarks           -v /raid/cuvs-bench-runner:/data/benchmarks           rapidsai/cuvs-bench:26.02-cuda13-py3.12 -c "python -m cuvs_bench.generate_groundtruth         /data/benchmarks/datasets/BeIR-trec-news-generated-queries-1.5M/base.fbin         --queries=/data/benchmarks/datasets/BeIR-trec-news-generated-queries-1.5M/query.fbin         --output /data/benchmarks/datasets/BeIR-trec-news-generated-queries-1.5M -k 10
        "
          
Reading whole dataset
Dataset size    4.6 GB, shape (1500000, 768), dtype float32
Reading queries from file /data/benchmarks/datasets/BeIR-trec-news-generated-queries-1.5M/query.fbin
Calculating true nearest neighbors
Step 0/3:
Step 1/3:
Step 2/3:
writing /data/benchmarks/datasets/BeIR-trec-news-generated-queries-1.5M/groundtruth.neighbors.ibin (10000, 10) uint32 ...
writing /data/benchmarks/datasets/BeIR-trec-news-generated-queries-1.5M/groundtruth.distances.fbin (10000, 10) float32 ...

Start with cuvs_bench.generate_groundtruth/main.py, especially import_with_fallback, and compare the 26.04 and 26.02 CUDA 13 Python 3.12 images. Reproduce the provided docker command and trace the CuPy CUDA-path failure. Done means the 26.04 GPU command computes and writes the ground-truth files as shown for 26.02.