Separate usage of Acc inside and outside of kernels

[BenjaminW3]

• Within a kernel it should probably be called context CtxGpuCuda& ctx so that it is easier for users to understand what it is
• Outside of a kernel it represents a stack of interconnected classes representing a backend

[psychocoderHPC]

I suggest a generic name e.g. DeviceCtx const & ctx

[sbastrakov]

Do we need device as part of name in kernel? - All kernels are always on device.

[BenjaminW3]

In the end it is about the variable name within the kernel because the type is already a template parameter. At the moment we have TAcc const & acc. I would simply replace it by TCtx const & ctx.

[ax3l]

I would like to note that for any non-computer scientist, the term "context" is not self-explanatory. So please document it well and prominently, if you pick it.

[sbastrakov]

@ax3l good point, but do you think accelerator is really more clear (not saying it's not, since I can't abstract from computer science background here)? A couple of weeks ago Rene, Jan, I and Michael struggled for some time in how to explain in words what does it mean exactly in terms of alpaka kernels, and it turned out we had at least 3 different opinions on that.

[ax3l]

Maybe DeviceContext or KernelContext or BackendInfo can add more meaning. As an idea, you five could focus on the description of what it is first and document it, and then a new name for the current accelerator type/param will arise more naturally.

[psychocoderHPC]

Since we use already the name work division we could also use WorkerCtx.

The definition we found on a discussion with @j-stephan, @sbastrakov , @bussmann was:

Abstraction of devices via a programming model. 
Maps your working division to the hardware cores via a given programming model, 
provides memory management and synchronization functionality depending
on that model and functions for computation.