simd: add simd_size_type

[ldh4]

Related to #5674 and #6109 (comment).

Introduced an alias, simd_size_type, for a signed int type.
Replaced size_t and int uses by simd_size_type as described in p1928r7 (section 4.10):

Different to the TS, this paper uses simd-size-type instead of size_t for
• the SIMD width (number of elements),
• the generator constructor call argument,
• the subscript operator arguments, and
...

[nliber]

#6109 (comment) was asking for a simd_size_type to be a private type. This change makes it a public type. Is that intentional?

I guess I'm asking whether or not this should be in Kokkos::Impl (or equivalent) or not.

I don't know that the answer is obvious one way or the other. For instance, simd_mask<T>::size() right now returns it, which would argue in favor of it being public. (If it were private, then simd_mask<T> ought to have a typedef for it.)

[ldh4]

#6109 (comment) was asking for a simd_size_type to be a private type. This change makes it a public type. Is that intentional?

I guess I'm asking whether or not this should be in Kokkos::Impl (or equivalent) or not.

I don't know that the answer is obvious one way or the other. For instance, simd_mask<T>::size() right now returns it, which would argue in favor of it being public. (If it were private, then simd_mask<T> ought to have a typedef for it.)

It was also unclear to me whether placing simd_size_type in an impl namespace provides a value. As you described, there will need an alias for Impl::simd_size_type to be used publicly in all backends. (Or using simd_size_type = Impl::internal_simd_size_type, if we want to preserve the names as consistently as possible with the synopsis provided in p1928r7 section 7.1.2)