Make `cuda::aligned_size_t` available in a more appropriate header than `<cuda/barrier>`

[jrhemstad]

The cuda::aligned_size_t<N> type is currently defined in <cuda/std/barrier>.

This requires me to include <cuda/std/barrier> any time I wish to use cuda::aligned_size_t. This is especially problematic as merely including cuda/barrier prevents compiling with <sm_70.

cuda::aligned_size_t is useful independent of its usage with barrier.

I want to be able to use cuda::aligned_size_t without including <cuda(/std)/barrier>.

I'm thinking we could add it to <cuda/cstddef>?

[sleeepyjack]

Could this potentially land in v1.9.0?

[jrhemstad]

You'd have to convince @wmaxey.

It is a relatively simple change, so I'd be fine with sneaking it in if we can, but Wesley calls the shots 'round here.

[sleeepyjack]

@wmaxey could this be a candidate for v1.9.0?

Also related: #163