examples: disable qwen3 thinking output in llama-swap requests

qwen3-8b emits <think>...</think> chain-of-thought before the JSON
response by default, which trips the JSON parser and forces ~50% of
posts to escalate to claude-agent-service (discovered during first
bulk ingest, 2026-06-05).

Fix: pass chat_template_kwargs.enable_thinking=false in the request
body for the Tier 1 (llama-swap) call. Claude calls are unaffected.

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

fire_planner/examples/llm_extract.py

@@ -62,6 +62,8 @@ async def extract_with_qwen(
         post=post,
         client=client,
         record_model=QWEN_MODEL,
+        # Suppress qwen3's chain-of-thought prefix so the response is bare JSON.
+        extra_body={"chat_template_kwargs": {"enable_thinking": False}},
     )
 
 
@@ -73,8 +75,9 @@ async def _call_openai_chat(
     client: httpx.AsyncClient,
     record_model: str,
     extra_headers: dict[str, str] | None = None,
+    extra_body: dict[str, Any] | None = None,
 ) -> ExtractedExample | None:
-    body = {
+    body: dict[str, Any] = {
         "model": model_name,
         "messages": [
             {"role": "system", "content": PROMPT_SYSTEM},
@@ -83,6 +86,8 @@ async def _call_openai_chat(
         "temperature": 0.0,
         "max_tokens": 512,
     }
+    if extra_body:
+        body.update(extra_body)
     try:
         resp = await client.post(
             url,