Simplify using NeighborSampler

[NST-d]

I currently have a usecase where i need to create batches from neighbors, but can't use NeighborLoader due to constant updates to the Data object.

While using NeighborLoader is very much plug and play, NeighborSampler seems to require doing all steps manually. Currently it looks something like:

sampler = NeighborSampler(graph, num_neighbors=num_neighbors)
starting_node = torch.randint(graph.num_nodes, (batch_size,))
input = NodeSamplerInput(node = starting_node, input_id=None)
output = sampler.sample_from_nodes(input)

batch_graph = graph.subgraph(output.node) 

Maybe using subgraph is also not optimal?

I suggest that there could be a util function which implements this behavior (or maybe an extension of NeighborSampler).