Fix summary LLM call: use raw chat text, not classifier-JSON parsing

client._chat() JSON-parses every response (for the classifier), so the plain-text
summary was rejected ("model did not return JSON") even though the model returned
a perfect summary. Split out _raw_content() and add chat_text() for free-form
output; summaries use it. _chat keeps parsing for classification.

goodnews/llm.py

@@ -206,7 +206,15 @@ class LocalModelClient:
                 names.append(str(model["id"]))
         return names
 
-    def _chat(self, payload: dict) -> dict:
+    def chat_text(self, messages: list[dict]) -> str:
+        """Plain chat completion → the raw message text (no JSON parsing).
+
+        Used for free-form output like summaries; classification uses _chat,
+        which JSON-parses the same content.
+        """
+        return self._raw_content(self._build_payload(messages, None))
+
+    def _raw_content(self, payload: dict) -> str:
         body = json.dumps(payload).encode("utf-8")
         headers = {"Content-Type": "application/json"}
         if self.api_key:
@@ -227,10 +235,12 @@ class LocalModelClient:
             raise RuntimeError(f"could not reach local model at {self.base_url}: {exc.reason}") from exc
 
         try:
-            content = data["choices"][0]["message"]["content"]
+            return data["choices"][0]["message"]["content"]
         except (KeyError, IndexError, TypeError) as exc:
             raise RuntimeError(f"unexpected local model response: {data}") from exc
-        return parse_classifier_json(content)
+
+    def _chat(self, payload: dict) -> dict:
+        return parse_classifier_json(self._raw_content(payload))
 
 
 @dataclass