Use CuPy for arrays & CUDA Python for low-level CUDA functions (#4958)

Switch to using CuPy for array creation when needed. Can still pass these to Numba or use `__cuda_array_interface__` for low-level operations.

Also switch to CUDA Python for low-level CUDA functions calls in Python.

Authors:
  - https://github.com/jakirkham

Approvers:
  - Rick Ratzel (https://github.com/rlratzel)

URL: #4958

Author: jakirkham

python/cugraph/cugraph/layout/force_atlas2_wrapper.pyx

@@ -1,4 +1,4 @@
-# Copyright (c) 2020-2023, NVIDIA CORPORATION.
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
@@ -22,7 +22,7 @@ from cugraph.structure.graph_primtypes cimport *
 from libcpp cimport bool
 from libc.stdint cimport uintptr_t
 import cudf
-from numba import cuda
+import cupy as cp
 import numpy as np
 
 cdef extern from "cugraph/legacy/internals.hpp" namespace "cugraph::internals":
@@ -67,7 +67,7 @@ def force_atlas2(input_graph,
     cdef GraphCOOView[int,int,double] graph_double
 
     df = cudf.DataFrame()
-    df['vertex'] = cudf.Series(np.arange(num_verts, dtype=np.int32))
+    df['vertex'] = cudf.Series(cp.arange(num_verts, dtype=cp.int32))
 
     src = input_graph.edgelist.edgelist_df['src']
     dst = input_graph.edgelist.edgelist_df['dst']
@@ -103,12 +103,12 @@ def force_atlas2(input_graph,
 
 
     # We keep np.float32 as results for both cases
-    pos = cuda.device_array(
+    pos = cp.empty(
             (num_verts, 2),
             order="F",
-            dtype=np.float32)
+            dtype=cp.float32)
 
-    pos_ptr = pos.device_ctypes_pointer.value
+    pos_ptr = pos.data.ptr
 
     if input_graph.edgelist.weights \
             and input_graph.edgelist.edgelist_df['weights'].dtype == np.float64:

python/cugraph/cugraph/tests/data_store/test_gnn_feat_storage_wholegraph.py

@@ -15,7 +15,7 @@
 import numpy as np
 import os
 
-import numba.cuda
+from cuda.bindings import runtime as cuda_runtime
 
 from cugraph.gnn import FeatureStore
 
@@ -28,8 +28,10 @@
 
 
 def get_cudart_version():
-    major, minor = numba.cuda.runtime.get_version()
-    return major * 1000 + minor * 10
+    status, version = cuda_runtime.getLocalRuntimeVersion()
+    if status != cuda_runtime.cudaError_t.cudaSuccess:
+        raise RuntimeError(f"CUDA Error: {status}")
+    return version
 
 
 pytestmark = [