[Sema] Fix incorrect enum token namespace

[etienneb]

Summary

This patch fix the scoping of enum literal. They were not resolving
to the right type.

It was not causing any problem as one is a copy of the other one.

The literal in the switch are resolving to Sema.h:5527

enum AccessResult {
  AR_accessible,
  AR_inaccessible,
  AR_dependent,
  AR_delayed
};

Instead of SemaAccess.cpp:27

/// A copy of Sema's enum without AR_delayed.
enum AccessResult {
  AR_accessible,
  AR_inaccessible,
  AR_dependent
};

This issue was found by a new clang-tidy check (still on-going).

[aaron.ballman]

LGTM!

lib/Sema/SemaAccess.cpp

Sema::AccessResult Sema::CheckFriendAccess(NamedDecl *target) {
   entity.setDiag(diag::err_access_friend_function)
       << (method->getQualifier() ? method->getQualifierLoc().getSourceRange()
                                  : method->getNameInfo().getSourceRange());
 
   // We need to bypass delayed-diagnostics because we might be called
   // while the ParsingDeclarator is active.
   EffectiveContext EC(CurContext);
   switch (CheckEffectiveAccess(*this, EC, target->getLocation(), entity)) {
-  case AR_accessible: return Sema::AR_accessible;
-  case AR_inaccessible: return Sema::AR_inaccessible;
-  case AR_dependent: return Sema::AR_dependent;
+  case ::AR_accessible: return Sema::AR_accessible;
+  case ::AR_inaccessible: return Sema::AR_inaccessible;
+  case ::AR_dependent: return Sema::AR_dependent;
   }
   llvm_unreachable("invalid access result");
 }
 
 Sema::AccessResult Sema::CheckAddressOfMemberAccess(Expr *OvlExpr,
                                                     DeclAccessPair Found) {
   if (!getLangOpts().AccessControl ||
       Found.getAccess() == AS_none ||