<type_traits>: Workaround for common_reference

[frederick-vs-ja]

Fixes #2581

Adds workaround for these cases:
When I is a cv-unqualified scalar type, LRef is cvI&, RRef is cvI&&, MSVC generally erroneously computes _Cond_res<LRef, RRef> as I&, while the right result is I. Although sometimes pointer types get right results, and it seems that I&& doesn't trigger the bug.

_Cond_res<volatile int&, const int&> is not fixed yet, but std::common_reference_t<volatile int&, const int&> doesn't rely on its result.

Additional workarounds:

• MSVC also sometimes computes _Cond_res<A&, A&&> incorrectly to A& when A is an array type. The result should be decayed.
• MSVC computes the composite pointer of volatile T(*)[N] and const T(*)[N] to ill-defined result type T volatile (const volatile*)[N]. Currently my workaround is incomplete. A complete workaround may need to hack the whole computation in [conv.qual]/3.
• DevCom-1627396 matters.

[StephanTLavavej]

Thanks! I went ahead and pushed changes for the really minor things I noticed.

Generally, I am allergic to permanently working around compiler deficiencies in type traits (as our ancient implementation of is_pod attempted to do) - that's bad for throughput, it's hard to know if the workaround is truly complete, and it accretes complexity as the language evolves. However, a compiler-specific, well-commented, well-tested, temporary workaround to a fundamental type trait that has been blocking a number of libcxx tests from running, seems like an entirely reasonable thing to do. (It's consistent with our usual practice of working around compiler bugs while reporting them and waiting for fixes.)

[CaseyCarter]

I'm mildly concerned that we'll discover some corner where trait(s) that are intended to reflect language behavior disagree with what the compiler actually does. Defining is_convertible_v<int, int(*)(float)> to true, for example, doesn't magically make that conversion work. "Lying" about reflection doesn't really fix a bug so much as relocate it. I can find no such issues here with a bit of cursory investigation, however, so I think it's safe to merge this.

[StephanTLavavej]

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

[StephanTLavavej]

Thanks for improving this type traits machinery to be more correct than the compiler! 😹 🚀 🎉