Generate skeleton kernels for the upper boundary of the domain

Which play well with the latest change to not assemble on the overlap.

python/dune/perftool/sumfact/switch.py

@@ -58,8 +58,11 @@ def decide_if_kernel_is_necessary(facedir_s, facemod_s, facedir_n, facemod_n):
 
     # The PDELab machineries visit-once policy combined with Yasp avoids any visits
     # with facemod_s being True
-    if facemod_s:
-        return False
+    # NB: This is not true anymore for parallel computations, as we would like to
+    #     skip computations on the overlap and that requires us to visit intersections
+    #     on the right/upper part of the domain from within the domain.
+    #if facemod_s:
+    #    return False
 
     # A codim1 entity can never be on the upper resp. lower side of the ref element
     # in both inside and outside cell in a YaspGrid