implement sorting, including argsort

[haydnv]

The fastest general-case parallel sort is IBR (interval-based rearrangement) bi-tonic: https://arxiv.org/abs/1511.03404

Simple bi-tonic sort and IBR bi-tonic sort performance converges around 2^17-2^20 keys

[haydnv]

This requires:

• slice: slice a Matrix #2
• concat: implement support for array concatenation #3
• write: https://docs.rs/custos/0.5.0/custos/trait.WriteBuf.html#tymethod.write

Sorting a key-value set (e.g. a (Matrix, Buffer) tuple of coordinates and values in a sparse tensor) will also require an argsort op which sorts a range of indices (for the Buffer according to the order of the keys in the coordinate Matrix)