fix: gateway proxy reads delta.message and detects truncated SSE

Standardize the stream delta on LangChain's own message dict (delta.message)
instead of a bespoke {text, tool_calls} shape, and raise when the stream ends
without the [DONE] sentinel so mid-stream truncation is no longer silent.

EvoScientist/llm/gateway_proxy.py

@@ -122,11 +122,13 @@ class GatewayProxyChatModel(BaseChatModel):
                 json=payload,
             ) as response:
                 response.raise_for_status()
+                saw_done = False
                 async for line in response.aiter_lines():
                     if not line.startswith("data:"):
                         continue
                     data = line[len("data:"):].strip()
                     if data == "[DONE]":
+                        saw_done = True
                         break
                     chunk = json.loads(data)
                     message = _chunk_to_message(chunk)
@@ -134,13 +136,18 @@ class GatewayProxyChatModel(BaseChatModel):
                         message=message,
                         generation_info=chunk.get("generation_info"),
                     )
+                if not saw_done:
+                    raise RuntimeError("AI4SCI_MODEL_STREAM_INCOMPLETE")
 
 
 def _chunk_to_message(chunk: dict[str, Any]) -> BaseMessage:
     delta = chunk.get("delta") or {}
-    parsed = messages_from_dict(
-        [delta.get("message") or {"type": "AIMessageChunk", "data": {"content": delta.get("text", "")}}]
-    )
+    message_dict = delta.get("message")
+    if message_dict is None:
+        raise RuntimeError("AI4SCI_MODEL_STREAM_DELTA_INVALID")
+    parsed = messages_from_dict([message_dict])
+    if len(parsed) != 1:
+        raise RuntimeError("AI4SCI_MODEL_STREAM_DELTA_INVALID")
     return parsed[0]