Fix tests on big-endian systems (#9380)

There is nothing in the data created in these tests that requires them
to be little endian, so the dtype comparison should be for native byte
order.

xarray/tests/test_backends.py

@@ -872,7 +872,7 @@ def test_roundtrip_empty_vlen_string_array(self) -> None:
                 if actual["a"].dtype.metadata is not None:
                     assert check_vlen_dtype(actual["a"].dtype) is str
             else:
-                assert actual["a"].dtype == np.dtype("<U1")
+                assert actual["a"].dtype == np.dtype("=U1")
 
     @pytest.mark.parametrize(
         "decoded_fn, encoded_fn",
@@ -1430,8 +1430,8 @@ def test_encoding_kwarg_vlen_string(
         expected = Dataset({"x": expected_string})
         kwargs = dict(encoding={"x": {"dtype": str}})
         with self.roundtrip(original, save_kwargs=kwargs) as actual:
-            assert actual["x"].encoding["dtype"] == "<U3"
-            assert actual["x"].dtype == "<U3"
+            assert actual["x"].encoding["dtype"] == "=U3"
+            assert actual["x"].dtype == "=U3"
             assert_identical(actual, expected)
 
     @pytest.mark.parametrize("fill_value", ["XXX", "", "bár"])

xarray/tests/test_coding_times.py

@@ -1253,7 +1253,7 @@ def test_roundtrip_datetime64_nanosecond_precision(
         encoding = {}
 
     var = Variable(["time"], times, encoding=encoding)
-    assert var.dtype == np.dtype("<M8[ns]")
+    assert var.dtype == np.dtype("=M8[ns]")
 
     encoded_var = conventions.encode_cf_variable(var)
     assert (
@@ -1264,7 +1264,7 @@ def test_roundtrip_datetime64_nanosecond_precision(
     assert encoded_var.data.dtype == dtype
 
     decoded_var = conventions.decode_cf_variable("foo", encoded_var)
-    assert decoded_var.dtype == np.dtype("<M8[ns]")
+    assert decoded_var.dtype == np.dtype("=M8[ns]")
     assert (
         decoded_var.encoding["units"]
         == f"{_numpy_to_netcdf_timeunit(timeunit)} since 1970-01-01 00:00:00"

xarray/tests/test_dask.py

@@ -1381,7 +1381,7 @@ def test_map_blocks_roundtrip_string_index():
     ds = xr.Dataset(
         {"data": (["label"], [1, 2, 3])}, coords={"label": ["foo", "bar", "baz"]}
     ).chunk(label=1)
-    assert ds.label.dtype == np.dtype("<U3")
+    assert ds.label.dtype == np.dtype("=U3")
 
     mapped = ds.map_blocks(lambda x: x, template=ds)
     assert mapped.label.dtype == ds.label.dtype