Add fixed-sized types to type system

[mplegendre]

Adiak currently uses int/long/unsigned long in its types. We should add the fixed-sized types:

int32_t, uint64_t, ....

We've also been asked to include the 128-bit types for future proofing.

For floating, points we could add 32-bit floats.
We might also consider the GPU-sized floating point: 8-bit, 16-bit, etc. Alternatively, should these types just be cast to float/double for the purposes of putting them into adiak?

[baallan]

I would advocate using an explicit enum and primitivee types for the smaller float sizes and not expect plugins or adiak to assume larger sizes for the adiak api. Functions in adiak (if any) that need to understand a particular vendor-specific float size are only provided (or only work) if adiak is compiled with a tool that supports that size. Casting leads to errors around arrays of smaller size floats and even just scalars if passed through a void*. Plugins can then make policy specific to themselves and their customers.