fix: replace_fn.cpp

Changes:

- We avoid the thread local storage.
- We use a hash map to ensure that cached values are not lost.
- We remove `check_system`. If this becomes an issue in the future we
should precompute the remaining amount of stack space, and use a cheaper
check.
- We add a `Expr.replaceImpl`, and will use it to implement
`Expr.replace` after update-stage0

src/kernel/for_each_fn.cpp

@@ -68,7 +68,7 @@ template<bool partial_apps> class for_each_fn {
             if (partial_apps)
                 apply(app_fn(e));
             else
-                apply_fn(e);
+                apply_fn(app_fn(e));
             apply(app_arg(e));
             return;
         case expr_kind::Lambda: case expr_kind::Pi:

src/kernel/replace_fn.cpp

@@ -105,7 +105,7 @@ class replace_fn {
         lean_inc_ref(m_f);
         lean_object * r = lean_apply_1(m_f, e.raw());
         if (!lean_is_scalar(r)) {
-            expr e_new(lean_ctor_get(r, 0));
+            expr e_new(lean_ctor_get(r, 0), true);
             lean_dec_ref(r);
             return save_result(e, e_new, shared);
         }