Add initial support of distributed sampling

[chang-l]

This PR introduces support for distributed graph sampling (via NCCL backend). The initial implementation focuses on the uniform neighbor sampler.
We are going to extend it to support other samplers in future.

Highlights:

• Distributed Graph Storage: Now, the graph structure (represented by row_ptr and col_indx tensors) can be stored as wholememory arrays in a distributed fashion with even distribution across ranks (support both cpu and cuda storage type).
• Distributed Sampling: The sampling process leverages the existing wholegraph gather function to collect the sampled nodes and edges across all ranks.
• Uniform Neighbor Sampler Support: Currently, only the uniform neighbor sampler is supported.

cc. @linhu-nv @dongxuy04 @BradReesWork @nvcastet @TristonC

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[linhu-nv]

It seems good to me.

[BradReesWork]

/okay to test

[BradReesWork]

/okay to test

[BradReesWork]

/merge