[FEA] Support getting all the neighboring edges for the uniform_neighbor_sample

[VibhuJawa]

Describe the solution you'd like and any additional context

For GNN workflows when running on validation/test we need to sample all the neighboring edges . I think we should support uniform_neighbor_sample to allow an option to return that.

If we want to match DGL behavior we can probably return all neighbors for Fan-Out=-1 .

See related DGL documentation here

Please note that currently below fails with CUGRAPH_UNKNOWN_ERROR , we should probably raise an informative error as -1 is the default value for DGL library and users may end up trying it on their own.

import cudf
import cugraph
import numpy as np

df = cudf.DataFrame({'source':[1,2,3,4],
                     'destination':[2,3,4,1]
                    })
df = df.astype(np.int32)
df['weight']=1.0

G = cugraph.Graph(directed=True)
G.from_cudf_edgelist(df,edge_attr='weight')

sample_df = cugraph.uniform_neighbor_sample(
    G,
    cudf.Series([1,2]),
    fanout_vals=[-1]
)

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
Input In [6], in <cell line: 14>()
     11 G = cugraph.Graph(directed=True)
     12 G.from_cudf_edgelist(df,edge_attr='weight')
---> 14 sample_df = cugraph.uniform_neighbor_sample(
     15     G,
     16     cudf.Series([1,2]),
     17     fanout_vals=[-1]
     18 )

File ~/dgl/cugraph/python/cugraph/cugraph/sampling/uniform_neighbor_sample.py:113, in uniform_neighbor_sample(G, start_list, fanout_vals, with_replacement, is_edge_ids)
    107 do_expensive_check = False
    109 sg = SGGraph(resource_handle, graph_props, srcs, dsts, weights,
    110              store_transposed, renumber, do_expensive_check)
    112 sources, destinations, indices = \
--> 113     pylibcugraph_uniform_neighbor_sample(resource_handle, sg, start_list,
    114                                          fanout_vals, with_replacement,
    115                                          do_expensive_check)
    117 df = cudf.DataFrame()
    118 df["sources"] = sources

File pylibcugraph/uniform_neighbor_sample.pyx:142, in pylibcugraph.uniform_neighbor_sample.uniform_neighbor_sample()

File pylibcugraph/utils.pyx:51, in pylibcugraph.utils.assert_success()

RuntimeError: non-success value returned from cugraph_uniform_neighbor_sample: CUGRAPH_UNKNOWN_ERROR

[ChuckHastings]

I believe that this should work. The fact that it fails is likely a bug.

There is a more detailed error message. We need to refactor the cuGraph python layer (or perhaps the python layer, not sure which) to expose the most detailed message. I will try and reproduce this specific example to isolate what is going on.