Test Deep Learning Frameworks in nightlies

@caryr35 and I were discussing and we would like to add automated nightly tests that spot-check RAPIDS + Deep Learning Frameworks.

Currently there is a manual process for running smoke tests. It is mostly imports and ensuring GPU support is available.

We can automate this process with CI for our nightlies.

PyTorch

First, an environment is created:

conda create -n rapids-25.08 -c rapidsai -c conda-forge -c nvidia rapids=25.08 python=3.13 'cuda-version>=12.0,<=12.9' 'pytorch=*=*cuda*'

Then the following smoke tests are run:

import cudf
import cuml
import cugraph
import torch
assert torch.cuda.is_available()

TensorFlow

First, an environment is created:
(Note: TensorFlow doesn't support Python 3.13 at the time of writing)

conda create -n rapids-25.08 -c rapidsai -c conda-forge -c nvidia rapids=25.08 python=3.12 'cuda-version>=12.0,<=12.9' tensorflow

Then the following smoke tests are run (modified slightly to assert rather than print):

import cudf
import cuml
import cugraph
import tensorflow as tf
gpus = tf.config.list_physical_devices("GPU")
print(gpus)
assert len(gpus) > 1

The issue names no files or existing CI entry points. Start by locating the repository's nightly CI workflows and current smoke-test automation, then reproduce the documented conda environments. Done means nightly jobs create the PyTorch and TensorFlow environments, import cudf, cuml, cugraph, and the framework, and verify GPU availability with the stated assertions.