Investigate arkouda.random.Generator.shuffle() function performance

[ajpotts]

A customer reported : I have been using the arkouda.random.Generator.shuffle() function and I notice its been a bottleneck in my computational speed, and it has not scaled up well with nnz or compute nodes.

@bengionz

[ajpotts]

From @e-kayrakli on slack:
On a quick look, Chapel’s randomStream.shuffle is serial, and Arkouda’s is built on top of that. A relatively quick, but memory-heavy, solution could be to use fillRandom to create a random index mapping array, and than move data in parallel, ideally using aggregators, based off of that. That is, Arkouda’s shuffle could use a server implementation like that.

[ajpotts]

One option would be distributed MergeSort: https://arxiv.org/pdf/1508.03167