fix: restore streaming in call_llama — heartbeat proxy requires stream=True

Without stream=True, llama-server returns plain JSON instead of SSE.
The heartbeat proxy can't send keepalive pings → Cloudflare 524.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

app.py

@@ -150,13 +150,31 @@ def call_llama(img_b64: str, prompt: str) -> str:
         }
     elif OUTPUT_MODE == "line":
         payload["grammar"] = LINE_GRAMMAR
+    payload["stream"] = True
     r = requests.post(
         f"{LLAMA_SERVER}/v1/chat/completions",
         json=payload,
         timeout=TIMEOUT,
+        stream=True,
     )
     r.raise_for_status()
-    return r.json()["choices"][0]["message"]["content"]
+    content = []
+    for line in r.iter_lines():
+        if not line:
+            continue
+        line = line.decode("utf-8") if isinstance(line, bytes) else line
+        if line.startswith("data: "):
+            data = line[6:]
+            if data.strip() == "[DONE]":
+                break
+            try:
+                chunk = json.loads(data)
+                delta = chunk["choices"][0]["delta"].get("content", "")
+                if delta:
+                    content.append(delta)
+            except (json.JSONDecodeError, KeyError, IndexError):
+                pass
+    return "".join(content)
 
 # ---------- Parsing ----------