fix(ollama): disable thinking for deepseek call sites

apps/api/tests/test_chat_manager_ollama_routing.py

@@ -103,6 +103,7 @@ async def test_nemoclaw_chat_uses_resolved_interactive_lane(
     url, payload = _FakeAsyncClient.posted[0]
     assert url == "http://gcp-a:11435/api/chat"
     assert payload["model"] == "deepseek-r1:14b"
+    assert payload["think"] is False
 
 
 def test_chat_manager_has_no_direct_gemini_generation_path() -> None:

apps/api/tests/test_decision_manager_ollama_routing.py

@@ -22,6 +22,7 @@ class _FakeResponse:
 
 class _FakeAsyncClient:
     posted_urls: list[str] = []
+    posted_payloads: list[dict[str, Any]] = []
     fail_urls: set[str] = set()
     response: str = ""
 
@@ -37,6 +38,7 @@ class _FakeAsyncClient:
 
     async def post(self, url: str, *, json: dict[str, Any]) -> _FakeResponse:
         self.posted_urls.append(url)
+        self.posted_payloads.append(json)
         if url in self.fail_urls:
             raise RuntimeError("endpoint unavailable")
         return _FakeResponse(self.response)
@@ -45,6 +47,7 @@ class _FakeAsyncClient:
 @pytest.fixture(autouse=True)
 def _reset_fake_client() -> None:
     _FakeAsyncClient.posted_urls = []
+    _FakeAsyncClient.posted_payloads = []
     _FakeAsyncClient.fail_urls = set()
     _FakeAsyncClient.response = ""
 
@@ -115,6 +118,7 @@ async def test_nemoclaw_second_opinion_tries_gcp_b_after_gcp_a_failure(
         "http://gcp-a:11435/api/generate",
         "http://gcp-b:11436/api/generate",
     ]
+    assert all(payload["think"] is False for payload in _FakeAsyncClient.posted_payloads)
 
 
 @pytest.mark.asyncio