Fix sklearn wrapper CI tests by marking pipeline consistency checks as expected failures

Neural networks are inherently non-deterministic, so pipeline consistency
checks should be skipped rather than fail. Added check_pipeline_consistency
to EXPECTED_FAILED_CHECKS for all sklearn wrapper types.

Author: amitsrivastava78

keras/src/wrappers/sklearn_test.py

@@ -107,16 +107,19 @@ def use_floatx(x):
         ),
         "check_supervised_y_2d": "This test assumes reproducibility in fit.",
         "check_fit_idempotent": "This test assumes reproducibility in fit.",
+        "check_pipeline_consistency": "Neural networks are non-deterministic",
     },
     "SKLearnRegressor": {
         "check_parameters_default_constructible": (
             "not an issue in sklearn>=1.6"
         ),
+        "check_pipeline_consistency": "Neural networks are non-deterministic",
     },
     "SKLearnTransformer": {
         "check_parameters_default_constructible": (
             "not an issue in sklearn>=1.6"
         ),
+        "check_pipeline_consistency": "Neural networks are non-deterministic",
     },
 }