[FEA]: P2655R3 common_reference_t of reference_wrapper Should Be a Reference Type

[davebayer]

This issue can be closed by implementing P2655R3

[HenrikGharagyozyan]

Hi! I'd like to take this one (I have two merged-track PRs in the repo, #10563/#10564). Plan: port libc++'s P2655R3 implementation — the __ref_wrap_common_reference_exists_with constraint plus the two mutually exclusive basic_common_reference partial specializations — into cuda/std/__functional/reference_wrapper.h, translated to CCCL's C++17-compatible machinery: the constraint as a _CCCL_CONCEPT, applied through basic_common_reference's fifth SFINAE parameter (same pattern as the existing tuple specialization) rather than native requires clauses, so it works across all supported dialects. The asymmetric exists_with(R,T) && !exists_with(T,R) condition prevents the wrapper-vs-wrapper self-ambiguity from the earlier paper revisions. Fully host-testable — I'll add static_assert coverage for the basic/mirrored cases, wrapper-vs-wrapper (ambiguity regression), const-correctness, and a negative case where no specialization may apply.

[davebayer]

Hi @HenrikGharagyozyan, I assigned you to the issue