<format> assumes strings are encoded in the active code page

[CaseyCarter]

Refactors all format functionality with knowledge of character encodings into a new class _Fmt_codec. _Fmt_codec internally caches the _Cvtvec structure that other functions previously passed around, and makes use of a new "pure C ABI function" __std_get_cvt to retrieve character conversion info for the active codepage.

The ABI function reuses the __std_code_page and __std_win_error types from <xfilesystem_abi.h>. These could be extracted into a more general __msvc_win_abi.hpp header, but I think <xfilesystem_abi.h> is lightweight enough to just include directly for now.

Drive-by:

• BUG: _Parse_precision was not skipping the trailing } in a dynamic precision.
• Improvement: test_parse_helper now verifies that the parse consumes the entire input when passed an expected length of npos.

[miscco]

The fact that you cooked that up in a few days makes my imposter syndrome super happy

[barcharcraz]

Looks pretty good to me, although the hack to hook into format.cpp to change the codepage needs to change.

I do observe that the ucrt uses the _setmbcp and _getmbcp functions, which seem to set a global (not great).

[CaseyCarter]

Looks pretty good to me, although the hack to hook into format.cpp to change the codepage needs to change.

No lie: this looked even jankier to me today than it did yesterday. I won't mourn its passing.

[StephanTLavavej]

Now that #1821 has merged, I've force-pushed and retargeted this to main. I inspected the history carefully - acp had accumulated 3 commits beyond feature/format. Because we took feature/format, added more commits to it in merging_format, and then squashed everything into main (squashing is awesome for linear history and I wouldn't give it up for anything, but it means that situations like this need extra care), I determined that the easiest and safest thing to do was to start from main and cherry-pick acp's 3 unique commits. This cherry-picked 99.9% cleanly, with the only conflict being this PR adding inclusion of <xfilesystem_abi.h> next to the removal of <variant>, which was trivial to resolve.

Apologies for the force-push! 😸

[StephanTLavavej]

Thanks for fixing this bug! 😻 🎉