Add stride constraint to XNN MaxPool (#7354)

Summary:
Add an XNNPACK partitioner constraint for MaxPool2d to enforce stride <= kernel_size.

See https://github.com/google/XNNPACK/blob/860f2b9ad9d3602599aff49a41d0131d2a350e00/src/subgraph/max-pooling-2d.c#L327.


Test Plan:
Added an operator-level test (test_fp32_maxpool2d_unsupported_stride) to verify the new constraint.

```
buck test executorch/backends/xnnpack/test:test_xnnpack_ops -- maxpool2d
```

Reviewed By: digantdesai

Differential Revision: D67380978

Pulled By: GregoryComer

Author: GregoryComer

backends/xnnpack/partition/config/generic_node_configs.py

@@ -282,15 +282,23 @@ class MaxPool2dConfig(GenericNodePartitionerConfig):
 
     def check_constraints(self, node: torch.fx.Node, ep: ExportedProgram) -> bool:
         """
-        XNNPACK's maxpool2d does not support ceil mode
+        XNNPACK's maxpool2d does not support ceil mode and requires stride <= kernel_size
         """
         if not self.check_common_constraints(node, ep):
             return False
 
+        kernel_size = node.args[1]
+        stride = node.args[2]
         is_ceil_mode = len(node.args) >= 6 and cast(bool, node.args[5])
+
         if is_ceil_mode:
             why(node, reason="ceil mode is not supported")
             return False
+
+        if stride[0] > kernel_size[0] or stride[1] > kernel_size[1]:
+            why(node, reason="stride must be less than or equal to kernel size")
+            return False
+
         return True
 
     def supported_precision_types(self) -> List[ConfigPrecisionType]:

backends/xnnpack/test/ops/test_maxpool2d.py

@@ -115,6 +115,28 @@ def test_fp32_maxpool2d_unsupported_ceilmode(self):
             .run_method_and_compare_outputs()
         )
 
+    def test_fp32_maxpool2d_unsupported_stride(self):
+        """
+        XNNPACK MaxPool2d requires stride <= kernel_size.
+        """
+        inputs = (torch.randn(1, 32, 23, 23),)
+        (
+            Tester(self.MaxPool2d(kernel_size=2, stride=3), inputs)
+            .export()
+            .check_count({"torch.ops.aten.max_pool2d.default": 1})
+            .to_edge_transform_and_lower()
+            # We expect it not be be delegated.
+            .check_count({"torch.ops.higher_order.executorch_call_delegate": 0})
+            .check_count(
+                {
+                    "executorch_exir_dialects_edge__ops_aten_max_pool2d_with_indices_default": 1
+                }
+            )
+            .to_executorch()
+            .serialize()
+            .run_method_and_compare_outputs()
+        )
+
     def test_qs8_maxpool2d(self):
         class MaxPool(torch.nn.Module):
             def __init__(self, maxpool_params):