feat(array-api): add cumulative_sum and cumulative_prod

These are the Array API standard equivalents of cumsum/cumprod with
three key differences that justify the separate names:

1. axis=None (default) flattens the input first; cumsum/cumprod require
   an explicit axis.
2. include_initial=True prepends the identity element (0 for sum, 1 for
   prod) so the output length along axis is len+1.  This matches the
   Array API spec's include_initial parameter and has no equivalent in
   cumsum/cumprod.
3. dtype parameter casts the input before accumulating, matching NumPy
   2.0 / Array API behaviour.

Docs and tests included.

Part of the array API split from ml-explore#3684.

Author: katlun-lgtm

docs/src/python/ops.rst

@@ -65,6 +65,8 @@ Operations
    cummin
    cumprod
    cumsum
+   cumulative_prod
+   cumulative_sum
    degrees
    depends
    dequantize

python/src/ops.cpp

@@ -5874,4 +5874,8 @@ void init_ops(nb::module_& m) {
   m.attr("empty_like") = m.attr("zeros_like");
   m.attr("matrix_transpose") = m.attr("transpose");
   m.attr("pow") = m.attr("power");
+  // Array API aliases — cumulative_sum/cumulative_prod are pure aliases of
+  // cumsum/cumprod, which now support dtype and include_initial.
+  m.attr("cumulative_sum") = m.attr("cumsum");
+  m.attr("cumulative_prod") = m.attr("cumprod");
 }

python/tests/test_ops.py

@@ -3446,6 +3446,28 @@ def test_to_from_fp8(self):
         self.assertTrue(mx.array_equal(mx.from_fp8(mx.to_fp8(vals)), vals))
         self.assertTrue(mx.array_equal(mx.from_fp8(mx.to_fp8(-vals)), -vals))
 
+    def test_cumulative_sum_prod(self):
+        a = mx.array([1, 2, 3, 4])
+        self.assertEqual(mx.cumulative_sum(a).tolist(), [1, 3, 6, 10])
+        self.assertEqual(
+            mx.cumulative_sum(a, include_initial=True).tolist(), [0, 1, 3, 6, 10]
+        )
+        self.assertEqual(mx.cumulative_prod(a).tolist(), [1, 2, 6, 24])
+        self.assertEqual(
+            mx.cumulative_prod(a, include_initial=True).tolist(), [1, 1, 2, 6, 24]
+        )
+
+        m = mx.array([[1, 2], [3, 4]])
+        self.assertEqual(mx.cumulative_sum(m, axis=0).tolist(), [[1, 2], [4, 6]])
+        self.assertEqual(mx.cumulative_sum(m, axis=1).tolist(), [[1, 3], [3, 7]])
+        self.assertEqual(
+            mx.cumulative_sum(m, axis=1, include_initial=True).tolist(),
+            [[0, 1, 3], [0, 3, 7]],
+        )
+        # axis=None flattens.
+        self.assertEqual(mx.cumulative_sum(m).tolist(), [1, 3, 6, 10])
+        self.assertEqual(mx.cumulative_sum(a, dtype=mx.float32).dtype, mx.float32)
+
 
 if __name__ == "__main__":
     mlx_tests.MLXTestRunner()