fix(backend): use json format for xgboost model marshalling (kubeflow#678)

Fixes kubeflow#675

Signed-off-by: Naymul Islam <naymul504@gmail.com>

Author: ai-naymul

examples/serving/xgboost/iris.ipynb

@@ -21,7 +21,9 @@
    "cell_type": "code",
    "execution_count": null,
    "metadata": {
-    "tags": []
+    "tags": [
+     "skip"
+    ]
    },
    "outputs": [],
    "source": [

kale/marshal/backends.py

@@ -110,7 +110,7 @@ class XGBoostModelBackend(MarshalBackend):
 
     name = "XGBoost Model backend"
     display_name = "xgboost"
-    file_type = "bst"
+    file_type = "json"
     obj_type_regex = r"xgboost\.core\.Booster"
     predictor_type = "xgboost"