O-16: distarray memory-mapped array functionality (optional)

Objective: To provide functionality that allows ODIN to use memory-mapped arrays as
the backing store for data.

Relevance: This provides support for users who have large storage capacity on, e.g. a workstation, and cannot fit an entire array in main memory. By allowing the backing store to use memory-mapped arrays, these users can perform distributed computations on arrays larger than can fit in main memory.

Description: The initial implementation of this task will likely build on top of NumPy’s existing memory-mapped array functionality. Based on this initial implementation and input from the distributed array protocol design, refinements will be made accordingly.