[clang] [Clang] Address feedback in PR183010 (PR #185608)

[Younan Zhang via cfe-commits]

================
@@ -1185,7 +1154,7 @@ static bool CheckConstraintSatisfaction(
     return false;
   }
 
-  if (TemplateArgsLists.isAnyArgInstantiationDependent(S.Context)) {
+  if (TemplateArgsLists.isAnyArgDependent(S.Context)) {
----------------
zyn0217 wrote:

> I think we can make the value judgment whether it's worth introducing a new regression to fix another regression, but I just want this to be acknowledged, FIXMEs exist for that reason.

@mizvekov 

We're not introducing a regression to fix another regressions. The issue you posted is caused by the normalization patch, which I think is at least expected given how normalization works.

Indeed I agree it's somewhat counter-intuitive because we seemed to accept `void &` from a user's perspective.

To move forward, how about adding your case to tests along with a FIXME there? Again isAnyArgDependent is not what caused this issue, so a FIXME here is confusing.

https://github.com/llvm/llvm-project/pull/185608