C++ binding updates for OpenCL 3.1

[bashbaug]

Updates the C++ bindings for OpenCL 3.1.

This mostly involved adding support for the new clGetKernelSuggestedLocalWorkSize with an associated new test, plus a bit of tidy up.

[Kerilk]

Looking good to me. Took me a bit of time to get why local is initialized to global (you want the same dimentionality, so the copy constructor gives you that). I think I would prefer it to be initialized to zero, but at the right dimentionality, but that is a refactoring of the class that may be out of scope of this PR.

[bashbaug]

Took me a bit of time to get why local is initialized to global [...]

Yeah, I couldn't think of a way to initialize it to zero (or some other default value) without either expanding the NDRange class a bunch or needing a bunch of conditional code. Initializing to global mostly does what I want it to do, so I went with it for now.

I've added a comment to avoid possible confusion in the future. We can revisit if we ever decide to update the NDRange class.