Add back a more restrictive short-circuit

Summary:
This is safe as we never have two handles with different sys info that are reachable from each other.

I've deliberately put this as two operations - remove the optimisation and then add back a different form, as it makes it clear the optimisation isn't load bearing.

Reviewed By: yangdanny97

Differential Revision: D80084280

fbshipit-source-id: bffd40085d38f6126a1b683127a7d4942f7fe938

Author: ndmitchell

pyrefly/lib/alt/answers_solver.rs

@@ -615,18 +615,14 @@ impl<'a, Ans: LookupAnswer> AnswersSolver<'a, Ans> {
         BindingTable: TableKeyed<K, Value = BindingEntry<K>>,
         SolutionsTable: TableKeyed<K, Value = SolutionsEntry<K>>,
     {
-        /*
-        // This is a tempting optimisation, but it is not sound as from a `py` file you can import your `pyi` file.
-        if module == self.module().name() && path.is_none_or(|path| path == self.module().path())
-        {
+        if module == self.module().name() && path == Some(self.module().path()) {
             // We are working in our own module, so don't have to go back to the `LookupAnswer` trait.
             // But even though we are looking at our own module, we might be using our own type via an import
             // from a mutually recursive module, so have to deal with key_to_idx finding nothing due to incremental.
             Some(self.get_idx(self.bindings().key_to_idx_hashed_opt(Hashed::new(k))?))
+        } else {
+            self.answers.get(module, path, k, self.thread_state)
         }
-        */
-
-        self.answers.get(module, path, k, self.thread_state)
     }
 
     pub fn get_from_export(