[simd.syn] Reorder declarations to match subclause order

[eisenwave]

Fixes NB US 175-281 (C++26 CD).

Fixes cplusplus/nbballot#856.

[eisenwave]

I also wrote down some thoughts on the issue at cplusplus/nbballot#856 (comment)

We could also reorder the subclauses, but in my opinion, they're structured very logically. All the basic_vec stuff first, then basic_mask. I don't see any compelling reason to deviate, and making the synopsis match the subclauses is the easiest fix.

@mattkretz can you confirm that the mismatch between the synopsis and the subclauses is accidental?

[mattkretz]

@mattkretz can you confirm that the mismatch between the synopsis and the subclauses is accidental?

The current order is derived from what got published in the TS. And the TS order was probably fairly accidental. Making the synopsis match the subclause ordering (rather than the other way around) seems like the safest thing to do.

FWIW, I implement basic_mask before basic_vec because that's just more natural in terms of code dependencies. The order in my implementation (not asking for a subclause reordering):

1. [simd.expos]
2. [simd.traits]
3. [simd.iterator]
4. [simd.creation] + [simd.permute.static]
5. [simd.mask] including all hidden friends in [simd.mask.nonmembers]
6. [simd.flags]
7. [simd.class] including all hidden friends in [simd.nonmembers]
8. [simd.loadstore]
9. remaining [simd.permute.*]
10. [simd.mask.reductions]
11. [simd.reductions]
12. [simd.alg]
13. [simd.bit]
14. complex specializations of [simd.mask] and [simd.class]
15. [simd.math]

[jensmaurer]

@tkoeppe , we need this to go in so that, as a next step, we can adjust the subclause headings quoted in the synopsis to the new reality, without causing merge conflicts.