NamedDefaults: implementation should not prevent tyvars appearing in non-unary constraints from being defaultable

GHC proposal #409 relaxed the rules for defaulting at use-sites. However, the implementation doesn't match the specification, as GHC.Tc.Solver.Default.findDefaultableGroups still considers any type variable that appears in a non-unary constraint to be non-defaultable.

So we should either:

1. Remove that check entirely, or
2. Remove that check when -XNamedDefaults is enabled. Note that this check is already skipped when -XExtendedDefaultRules is enabled.

(2) seems like the easier sell, but I still find it quite strange that enabling/disabling the extension would affect this case (non-unary constraints) when the rest of the logic (such as defaulting non-standard classes) is not gated at all (see my analysis in this comment).

See also

• #25775 (closed)
• #25807 (closed)