STL: `std::iter_swap` can be inlined into callers

[CaseyCarter]

After merging P0551R3 "Thou Shalt Not Specialize std Function Templates!" users are now forbidden to explicitly specialize std::iter_swap. Consequently, we can expand calls inline into callsites for a bit of throughput and improved debug codegen / UX under as-if. Should the expansion be _Swap_adl(*a, *b);, or should we inline that as well and use swap(*a, *b); // intentional ADL? _Swap_adl is conventional, but perhaps its time to update that convention for another tiny bit of throughput?

We should see what other maintainers think about inlining _Swap_adl before wasting any effort on doing so - I'll label this [decision needed] for now.

[frederick-vs-ja]

I think we can do this even without the strengthened requirement in P0551R3.

The old wording said...

N4140 [namespace.std]/1:

A program may add a template specialization for any standard library template to namespace std only if the declaration depends on a user-defined type and the specialization meets the standard library requirements for the original template and is not explicitly prohibited.

N4140 [alg.swap]/5:

Effects: swap(*a, *b).

According to the old wording, IIUC the previously allowed program-defined specializations of std::iter_swap were already disallowed to do something inequivalent.

[StephanTLavavej]

We talked about this at the weekly maintainer meeting, and everyone likes swap(*a, *b); // intentional ADL. (Let's make sure to consistently comment this - also IIRC there is at least one uncommented occurrence already somewhere in the STL.)

Rationale: This probably doesn't matter a whole bunch for throughput (either compile time or object file size) since the number of swap calls is fairly low, but it will improve the debugging experience to have to step through only one swap function instead of 2 extra layers.