Fix：The OpenAI-Compatible Agent API returns an incorrect message (#8177)

### What problem does this PR solve?

https://github.com/infiniflow/ragflow/issues/8175

### Type of change

- [x] Bug Fix (non-breaking change which fixes an issue)

agent/canvas.py

     def run(self, running_hint_text = "is running...🕞", **kwargs):
         if not running_hint_text or not isinstance(running_hint_text, str):
             running_hint_text = "is running...🕞"
+        bypass_begin = bool(kwargs.get("bypass_begin", False))
 
         if self.answer:
             cpn_id = self.answer[0]
         if not self.path:
             self.components["begin"]["obj"].run(self.history, **kwargs)
             self.path.append(["begin"])
+            if bypass_begin:
+                cpn = self.get_component("begin")
+                downstream = cpn["downstream"]
+                self.path.append(downstream)
+
+
 
         self.path.append([])
 

agent/component/answer.py

             for ii, row in stream.iterrows():
                 answer += row.to_dict()["content"]
                 yield {"content": answer}
-        else:
+        elif stream is not None:
             for st in stream():
                 res = st
                 yield st
-        if self._param.post_answers:
+        if self._param.post_answers and res:
             res["content"] += random.choice(self._param.post_answers)
             yield res
 
+        if res is None:
+            res = {"content": ""}
+
         self.set_output(res)
 
     def set_exception(self, e):

agent/component/base.py

             elif q.get("value"):
                 outs.append(pd.DataFrame([{"content": q["value"]}]))
         return outs
-
     def get_input(self):
         if self._param.debug_inputs:
             return pd.DataFrame([{"content": v["value"]} for v in self._param.debug_inputs if v.get("value")])

api/db/services/canvas_service.py

             "source": "agent",
             "dsl": cvs.dsl
         }
+        canvas.messages.append({"role": "user", "content": question, "id": message_id})
+        canvas.add_user_input(question)
+
         API4ConversationService.save(**conv)
         conv = API4Conversation(**conv)
+        if not conv.message:
+            conv.message = []
+        conv.message.append({
+            "role": "user",
+            "content": question,
+            "id": message_id
+        })
+        
+        if not conv.reference:
+            conv.reference = []
+        conv.reference.append({"chunks": [], "doc_aggs": []})
             
     # Handle existing session
     else:
     if stream:
         try:
             completion_tokens = 0
-            for ans in canvas.run(stream=True):
+            for ans in canvas.run(stream=True, bypass_begin=True):
                 if ans.get("running_status"):
                     completion_tokens += len(tiktokenenc.encode(ans.get("content", "")))
                     yield "data: " + json.dumps(
     else:  # Non-streaming mode
         try:
             all_answer_content = ""
-            for answer in canvas.run(stream=False):
+            for answer in canvas.run(stream=False, bypass_begin=True):
                 if answer.get("running_status"):
                     continue