Roadmap to 1.0

- #317
- Define KernelAbstractionsCore.jl that defines the interface for backend
- Move backends from CUDAKernels to CUDA and so-forth

The last two are necessary since I just noticed that if you write a library that just uses KA the user always needs to couple their choice of CUDA library with the KernelAbstractions backend.

```
using CUDA
using CUDAKernels
```

Start by reviewing issue #317 and the current KernelAbstractions and CUDAKernels structure. Investigate how KernelAbstractionsCore.jl would define the backend interface and how backends could move from CUDAKernels to CUDA and other packages; the roadmap is complete when the coupling between a library using KA and a user's chosen CUDA library is removed.

Written by the indexing model from the issue text.