add fake_quantize_per_channel_affine expected failure case for non-integer zero_point

Author: ruro

tests/function_libs/torch_lib/ops_test_data.py

@@ -700,6 +700,9 @@ def _where_input_wrangler(
     ),
     TorchLibOpInfo(
         "fake_quantize_per_channel_affine", core_ops.aten_fake_quantize_per_channel_affine
+    ).xfail(
+        reason="fixme: ONNX (De)QuantizeLinear only supports integer zero_point values",
+        matcher=lambda sample: sample.args[1].dtype != torch.int32,
     ),
     TorchLibOpInfo(
         "fake_quantize_per_tensor_affine", core_ops.aten_fake_quantize_per_tensor_affine