This should make it easier to get better output when wchar_t tests fail. The code is based on the Unicode transcoding in `<format>`. Differential Revision: https://reviews.llvm.org/D150593