Simplify (and fix) passing of guided decoding backend options (#17008)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>

examples/online_serving/openai_chat_completion_structured_outputs.py

@@ -112,8 +112,8 @@ def extra_backend_options_completion(client: OpenAI, model: str):
               "alan.turing@enigma.com\n")
 
     try:
-        # The no-fallback option forces vLLM to use xgrammar, so when it fails
-        # you get a 400 with the reason why
+        # The guided_decoding_disable_fallback option forces vLLM to use
+        # xgrammar, so when it fails you get a 400 with the reason why
         completion = client.chat.completions.create(
             model=model,
             messages=[{
@@ -123,7 +123,8 @@ def extra_backend_options_completion(client: OpenAI, model: str):
             extra_body={
                 "guided_regex": r"\w+@\w+\.com\n",
                 "stop": ["\n"],
-                "guided_decoding_backend": "xgrammar:no-fallback"
+                "guided_decoding_backend": "xgrammar",
+                "guided_decoding_disable_fallback": True,
             },
         )
         return completion.choices[0].message.content