Add use cases section

[rgommers]

Adds four concrete uses cases:

• add GPU and distributed support to SciPy
• simplify einops by removing the backend system
• adding a Python API to xtensor
• make JIT compilation of array computations easier and more robust

[shoyer]

This is great, thanks for adding this!

[shoyer]

I would consider the more generic term "hardware accelerator" rather than GPU here. There is a huge boom in AI accelerators at present (e.g., Google TPU + a many start-up products) and given current trends (i.e., the end of Moore's law) it seems increasingly likely that support for these devices will be an important part of scientific computing over the next decade.

[rgommers]

Good point. I think that's less on the radar of the average advanced end user so far, but it will certainly become of more interest soon. Adopted your suggestion.

[rgommers]

This seems good to go and has had two people look at it, so I'll plan on merging it within a day or two unless there are more comments.