Implement internal vector type traits

[davebayer]

The PR introduces interanl cuda::__vector_type_t and cuda::__has_vector_type_v traits for crating CUDA vector types from type and size.

I will need this type for hierarchy query result type.

[github-actions[bot]]

😬 CI Workflow Results

🟥 Finished in 1h 19m: Pass: 98%/88 | Total: 1d 00h | Max: 1h 17m | Hits: 99%/211903

See results here.

[davebayer]

I think most places where CubVector<...>::Type is used, we can use this new facility now.

Will it improve anything?

We may also want to revive some more parts of #4674.

I am not quite sure. Adding the _a16 and _a32 variants adds much more complexity to the design. I would go for the exclusive internal exposure only

[github-actions[bot]]

🥳 CI Workflow Results

🟩 Finished in 2h 23m: Pass: 100%/88 | Total: 23h 58m | Max: 1h 17m | Hits: 99%/211903

See results here.

[bernhardmgruber]

I think most places where CubVector<...>::Type is used, we can use this new facility now.

Will it improve anything?

Yes, because then I can deprecate and eventually remove CubVector<...>, or we replace it by something better.

[davebayer]

Yes, because then I can deprecate and eventually remove CubVector<...>, or we replace it by something better.

But the problem is that you cannot just implement the arithmetic operators for these types as CubVector does

[github-actions[bot]]

🥳 CI Workflow Results

🟩 Finished in 2h 08m: Pass: 100%/88 | Total: 1d 17h | Max: 2h 05m | Hits: 93%/213129

See results here.