[AIR][Serve] Add windows check for pd.DataFrame comparison #26889

n previous implementation #26821 we have windows failure suggesting we behave differently on windows regarding datatype conversion.

In our https://sourcegraph.com/github.com/ray-project/ray/-/blob/python/ray/data/tests/test_dataset.py?L577 regarding use of TensorArray we seem to rely on pd'sassert_frame_equal rather than manually comparing frames.

This PR adds a quick conditional on windows only to ignore dtype for now.

python/ray/serve/tests/test_air_integrations.py

@@ -1,3 +1,4 @@
+import os
 import tempfile
 from typing import Optional
 
@@ -90,8 +91,13 @@ def test_dataframe_with_tensorarray(self):
 
         unpacked_list = BatchingManager.split_dataframe(batched_df, 1)
         assert len(unpacked_list) == 1
-        assert unpacked_list[0]["a"].equals(split_df["a"])
-        assert unpacked_list[0]["b"].equals(split_df["b"])
+        # On windows, conversion dtype is not preserved.
+        check_dtype = not os.name == "nt"
+        pd.testing.assert_frame_equal(
+            unpacked_list[0].reset_index(drop=True),
+            split_df.reset_index(drop=True),
+            check_dtype=check_dtype,
+        )
 
 
 class AdderPredictor(Predictor):