Editable builds of `cuda-bench` are broken

[NaderAlAwar]

Following the installation instructions in our readme to build cuda-bench from source is broken

pip install -e .[cu12]

and running the python auto_throughput.py results in

Traceback (most recent call last):
  File "/home/nawar/projects/nvbench/python/cuda/bench/__init__.py", line 55, in <module>
    _nvbench_module = importlib.import_module(_module_fullname)
  File "/home/nawar/anaconda3/envs/python-nvbench/lib/python3.13/importlib/__init__.py", line 88, 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 1310, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 488, in _call_with_frames_removed
  File "<frozen importlib._bootstrap>", line 1387, in _gcd_import
  File "<frozen importlib._bootstrap>", line 1360, in _find_and_load
  File "<frozen importlib._bootstrap>", line 1324, in _find_and_load_unlocked
ModuleNotFoundError: No module named 'cuda.bench.cu12'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/home/nawar/projects/nvbench/python/examples/auto_throughput.py", line 19, in <module>
    import cuda.bench as bench
  File "/home/nawar/projects/nvbench/python/cuda/bench/__init__.py", line 57, in <module>
    raise ImportError(
    ...<4 lines>...
    )
ImportError: No cuda-bench extension found for CUDA 12.x. This wheel may not include support for your CUDA version. Supported CUDA versions: 12, 13. Original error: No module named 'cuda.bench.cu12'

This is due to #289, which added support for multiple CUDA versions, and moved the Python extension into versioned subpackages (e.g., cuda.bench.cu12._nvbench), so editable installs that don’t build/install that versioned layout fail to import the module.

(side note our README needs to be updated to specify [cu12] in the pip installation command)