<regex>: Compare against char directly

[Stephan T. Lavavej (StephanTLavavej)]

Fixes #5673 and fixes VSO-2548012 in conjunction with small Build2 changes.

Build2 is doing something extremely complicated and fragile with user-defined character types and regex. When we construct user-defined _Elem from ordinary chars, that triggers assertions in their machinery, even though the resulting homogeneous comparisons between _Elem and _Elem are arguably more cromulent.

This PR slightly backs off from the purity of such comparisons, resulting in code that's somewhat closer to our old way of doing things, but still avoiding the internal _Uelem machinery. Now we directly compare _Elem to char. Build2 has heterogeneous comparison operators that handle this.

It will still be necessary to update Build2 to round-trip their user-defined character type instead of asserting, but that's a less invasive change: build2/build2#478

This makes no difference to the vast, vast majority of users who simply use char or wchar_t.

[David Justo (davidmrdavid)]

LGTM, just some questions

[Julian Müller (muellerj2)]

Looks good to me as well.

I also briefly checked the remaining static_cast's to _Elem in <regex>, but these are either meaningful round-trip casts or they appear in instances where we can't avoid casting to _Elem at some point (at least as long as we don't want to support heterogeneous char_traits or regex_traits implementations, which even Build2 does not provide.)

[Stephan T. Lavavej (StephanTLavavej)]

David Justo (@davidmrdavid) Julian Müller (@muellerj2) I've pushed changes to the test, improving the casts and centralizing the operators. Thanks for the reviews! 😻

[Stephan T. Lavavej (StephanTLavavej)]

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

[Stephan T. Lavavej (StephanTLavavej)]

I've pushed a non-trivial merge with main, reverse-mirroring what I did in MSVC-PR-662263.

This fixes the conflict in GH_000995_regex_custom_char_types with #5671 (which had moved and altered code in the test). Now, signed_wchar_enum and ullong_enum are gaining comparison operators through the same technique, wrapped_character's operators have been adjusted now that it's templated on Elem, and they're all implemented in terms of each other. (I suppose I was ultimately persuaded that it's the simpler thing to do, thanks for the review comment.)