Fix support of sequence protocol for returned lists

Add the `sequence` option to PyO3's `pyclass`, so that the `sq_length`
slot is implemented [1]. Implementing this method is required for
sequence types, or Python C API functions like `PySequence_Size` will
fail with an error.

The `reversed` built-in method relies on `PySequence_*` methods. A test
reversing `NodeIndices` is added to guard against future violations of
the sequence protocol.

Fixes Qiskit#696.

[1]: PyO3/pyo3#2567

src/iterators.rs

@@ -475,7 +475,7 @@ impl PyConvertToPyArray for Vec<(usize, usize, PyObject)> {
 macro_rules! custom_vec_iter_impl {
     ($name:ident, $data:ident, $T:ty, $doc:literal) => {
         #[doc = $doc]
-        #[pyclass(module = "rustworkx")]
+        #[pyclass(module = "rustworkx", sequence)]
         #[derive(Clone)]
         pub struct $name {
             pub $data: Vec<$T>,

tests/rustworkx_tests/test_custom_return_types.py

@@ -174,6 +174,12 @@ def test_slices_negatives(self):
         self.assertEqual([2, 3], slice_return)
         self.assertEqual([], indices[-1:-2])
 
+    def test_reversed(self):
+        indices = self.dag.node_indices()
+        reversed_slice = indices[::-1]
+        reversed_elems = list(reversed(indices))
+        self.assertEqual(reversed_slice, reversed_elems)
+
     def test_numpy_conversion(self):
         res = self.dag.node_indexes()
         np.testing.assert_array_equal(np.asarray(res, dtype=np.uintp), np.array([0, 1]))