create an "adhoc" module for "artificial" nodes

the module is specifically for nodes that are not based in the real
   syntactic tree, but created "ad-hoc", for example, new namedtuple
   classes that we create (see brain_namedtuple_enum).

This is the base for future changes on replacing non-module
  roots (with the adhoc module or something more approriate).

astroid/brain/brain_builtin_inference.py

@@ -24,6 +24,7 @@
 from astroid.inference_tip import inference_tip
 from astroid.manager import AstroidManager
 from astroid.nodes import scoped_nodes
+from astroid.raw_building import build_module
 from astroid.typing import (
     ConstFactoryResult,
     InferenceResult,
@@ -164,6 +165,8 @@ def _extend_builtins(class_transforms):
 
 def on_bootstrap():
     """Called by astroid_bootstrapping()."""
+    AstroidManager().cache_module(build_module("__astroid_adhoc"))
+
     _extend_builtins(
         {
             "bytes": partial(_extend_string_class, code=BYTES_CLASS, rvalue="b''"),

astroid/manager.py

@@ -115,6 +115,10 @@ def unregister_transform(self):
     def builtins_module(self) -> nodes.Module:
         return self.astroid_cache["builtins"]
 
+    @property
+    def adhoc_module(self) -> nodes.Module:
+        return self.astroid_cache["__astroid_adhoc"]
+
     @property
     def prefer_stubs(self) -> bool:
         return AstroidManager.brain["prefer_stubs"]