Implement proposed resolution of LWG-3540

[CaseyCarter]

... which makes <format> handle void* in the same manner as const void*. (Previously void* was treated as a "custom type" by the format argument machinery.)

Drive-by: swap order of parameters for concept _Has_formatter so we can use it as a type-constraint. Don't use return-type deduction for overloads of _Get_format_arg_storage_type that always return the same type.

Note that I haven't annotated this as implementing an unmerged LWG issue. The resolution hasn't been at all controversial in LWG, so I feel safe just making the change without annotation; reviewers may feel differently.

[miscco]

Looks great

[eldakesh-ms]

Do we lose test coverage where void* was used to stimulate a custom formatter?

[CaseyCarter]

Do we lose test coverage where void* was used to stimulate a custom formatter?

I certainly hope we have no such fragile test coverage. grep tells me that void\s*\* occurs only in formatting_args and formatting_formatting, and none of those occurrences looks to me like it intends to test custom formatting.

[StephanTLavavej]

Thanks for implementing this issue's resolution faster than the Standard could accept it! 😹 🚀 🚀 🚀