<format>: Implement P2418R2 support for std::generator-like types

[barcharcraz]

Implements P2418R2 which prepares format to be able to format std::generators, by allowing "format" functions that take their parameters by non-const reference. This DR is somewhat high risk because it modifies the argument type mapping machinery.

• make(w)_format_args was modified to forward its arguments, I think this is what the standard says to do but I'm not 100% sure since the standard has make_format_args directly initialize each basic_format_arg while we need to pass the arguments to all together to _Format_arg_store.
• I have modified usages of _Storage_type to use remove_cvref_t<T> instead of T because that's what the "other end" (the un-erase from handle) uses now. I don't think this has actual effect on the mapping, but could change the selected formatter specialization for custom types. Tim Song indicated on the reflectors that the lack of remove_cvref_t<T> in [format.arg]/5 is probably a defect.
• The _Has_formatter and the new _Has_const_formatter concepts have been moved up since they are now needed in the constructor for basic_format_arg::handle

http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2021/p2418r2.html

Fixes #2239.

[CaseyCarter]

Why are we forwarding when the Working Draft doesn't show forwarding? (Also on 2966.) If we don't forward here, the _Format_arg_store constructor can/should be changed to take only lvalues.

Have you submitted an LWG issue to fix the preconditions on these functions to strip cvref-quals from the T_i?

[barcharcraz]

I'm forwarding because libfmt forwards here, I think you're right though and they forward everywhere but then specifically undo what the forwarding would have done when make_arg binds them all to the fmt equivalent to const _Storage_type<remove_cvref_t<T>>&

Have you submitted an LWG issue to fix the preconditions on these functions to strip cvref-quals from the Ti?

no, but Tim Song did.

[barcharcraz]

wait is this a separate issue from LWG-3631?

[CaseyCarter]

Yes it is - [format.arg.store]/2. We can do this as a followup - it doesn't need to block merging.

[StephanTLavavej]

I pushed a conflict-free merge with main, in addition to a comment change, because there was the major toolset update containing Clang 13. I verified that no additional reformatting was necessary.

[StephanTLavavej]

I'm mirroring this to the MSVC-internal repo - please notify me if any further changes are pushed.

[StephanTLavavej]

Thanks for implementing the last C++20 DR! 🎉 😻 ✅