[BoundsSafety] Work around a regression in how macro-defined bounds-safety attributes are emitted in diagnostics

A recent auto-merged change from upstream
d9ef62a
seemed to remove the special handling on how macro-defined bounds-safety
attributes are emitted in diagnostics. Adding the specialization for now
to fix the test failure. The task to clean up the special handling of
the macro defined attributes is tracked separately.

This fixes test failures in
  Clang :: BoundsSafety/Sema/attributes_in_template_decls_attr_only_mode.cpp
  Clang :: BoundsSafety/Sema/complex-typespecs-with-bounds.c
  Clang :: BoundsSafety/Sema/terminated-by-attr.c
  Clang :: BoundsSafety/Sema/unsafe-late-const.c

rdar://155652853

Author: rapidsna

clang/include/clang/Sema/ParsedAttr.h

@@ -1110,6 +1110,15 @@ enum AttributeDeclKind {
   ExpectedTypedef,
 };
 
+inline const StreamingDiagnostic &operator<<(const StreamingDiagnostic &DB,
+                                             const ParsedAttr &At) {
+  const IdentifierInfo *AttrName =
+      At.printMacroName() ? At.getMacroIdentifier() : At.getAttrName();
+  DB.AddTaggedVal(reinterpret_cast<uint64_t>(AttrName),
+                  DiagnosticsEngine::ak_identifierinfo);
+  return DB;
+}
+
 } // namespace clang
 
 #endif // LLVM_CLANG_SEMA_PARSEDATTR_H