[FEA] Support DGL NegativeSampler needed for Link Prediction

[VibhuJawa]

Describe the solution you'd like and any additional context

We should scope and test the negative sampling we do for DGL cugraph objects

https://docs.dgl.ai/en/0.9.x/guide/minibatch-link.html#guide-minibatch-link-classification-sampler

class NegativeSampler(object):
    def __init__(self, g, k):
        # caches the probability distribution
        self.weights = g.in_degrees().float() ** 0.75
        self.k = k

    def __call__(self, g, eids):
        src, _ = g.find_edges(eids)
        src = src.repeat_interleave(self.k)
        dst = self.weights.multinomial(len(src), replacement=True)
        return src, dst


sampler = dgl.dataloading.as_edge_prediction_sampler(
    sampler, negative_sampler=NegativeSampler(g, 5))
dataloader = dgl.dataloading.DataLoader(
    g, train_seeds, sampler,
    batch_size=args.batch_size,
    shuffle=True,
    drop_last=False,
    pin_memory=True,
    num_workers=args.num_workers)

Might be related:
https://docs.dgl.ai/en/0.9.x/_modules/dgl/sampling/negative.html#global_uniform_negative_sampling

---

For Developers below this line

• [] Code passes CI
• [] Code uses Graph Primitives
• [] Code in C++/CUDA layer
• [] Code in C layer
• [] Code in pylibcugraph Python layer
• [] Code in cugraph Python layer
• [] Documentation
• [] Test cases at each layer (for MG tests, a note in the PR description indicating the new/existing MG tests were run and passed)