Handle RuntimeError thrown by CUDA Python in validate_setup (#10653)

The call to `getDeviceCount()` can raise a `RuntimeError` when `libcuda.so` is missing. We should handle that too in `validate_setup()`.

Authors:
  - Ashwin Srinath (https://github.com/shwina)

Approvers:
  - Charles Blackmon-Luca (https://github.com/charlesbluca)
  - GALI PREM SAGAR (https://github.com/galipremsagar)

URL: #10653

python/cudf/cudf/utils/gpu_utils.py

@@ -55,6 +55,12 @@ def validate_setup():
             raise e
         # If there is no GPU detected, set `gpus_count` to -1
         gpus_count = -1
+    except RuntimeError as e:
+        # getDeviceCount() can raise a RuntimeError
+        # when ``libcuda.so`` is missing.
+        # We don't want this to propagate up to the user.
+        warnings.warn(str(e))
+        return
 
     if gpus_count > 0:
         # Cupy throws RunTimeException to get GPU count,