Add sample_edges and sample_nodes

[rogerpearce]

After #56 is merged. Create a PR off master to do this.

Goal is to add ability to uniformly random sample from a where clause's subgraph k nodes or edges. Key difference from select_* is that it uniformly samples.

To do this we must first compute the full where-subgraph, then compute a prefix_sum across the distribution of processors, then sample the k indices to be marked for returning. Assume k is rather small, under 1000. k should be a required parameter without a default.

Note that YGM already has support for prefix_sum:
https://github.com/llnl/ygm/blob/master/include/ygm/detail/collective.hpp#L25

[sbromberger]

I think this should add a boolean series that is true if the element is sampled. This way we can have large samples without worrying about running out of memory on the front end. If the samples are small, then a followup select_nodes or select_edges with an appropriate where clause would return the data.

This also simplifies the API significantly as we don't have to require an additional list of metadata for this command.