Fix width-one iteration slices on GPU

When iteration_slice was specified, the GPU code generator used to handle integer slice components wrongly, setting start == stop instead of stop == start + 1. This MR fixes that and provides tests.

Also, this reveals a more fundamental problem with iteration slices on GPU; see #103

tests/test_sliced_iteration.py

@@ -16,6 +16,9 @@ from pystencils.simp import sympy_cse_on_assignment_list
 
 @pytest.mark.parametrize("target", [Target.CPU, Target.GPU])
 def test_sliced_iteration(target):
+    if target == Target.GPU:
+        pytest.importorskip("cupy")
+
     size = (4, 4)
 
     dh = create_data_handling(size, default_target=target, default_ghost_layers=0)