perf(model-metadata): resolve localhost to IPv4 in detect_local_server_type

On Windows, `localhost` resolves to both ::1 (IPv6) and 127.0.0.1 (IPv4).
httpx tries IPv6 first, hanging 2 sec per probe when the server binds IPv4
only. detect_local_server_type() is called 3+ times during init, each with
a new httpx.Client, compounding to ~14s of dead time.

Replace localhost with 127.0.0.1 inside the function before connecting.
The function is only called for local endpoints (callers guard with
is_local_endpoint()), so IPv6 loopback adds no diagnostic value.

Measured: 19.9s → 4.0s on Windows with a local proxy on 127.0.0.1:8317.
(cherry picked from commit a075d3194b)

agent/model_metadata.py

@@ -648,10 +648,19 @@ def detect_local_server_type(base_url: str, api_key: str = "") -> Optional[str]:
     import httpx
 
     normalized = _normalize_base_url(base_url)
+
+    # Resolve localhost to IPv4 to avoid 2s IPv6 timeout on Windows dual-stack.
+    # Applied to ``normalized`` before deriving server/LM Studio URLs AND
+    # before the cache lookup, so localhost and 127.0.0.1 share a cache entry.
+    normalized = normalized.replace("://localhost:", "://127.0.0.1:")
+    normalized = normalized.replace("://localhost/", "://127.0.0.1/")
+    if normalized.endswith("://localhost"):
+        normalized = normalized[:-len("localhost")] + "127.0.0.1"
+
     server_url = normalized
     if server_url.endswith("/v1"):
         server_url = server_url[:-3]
-    lmstudio_url = _lmstudio_server_root(base_url)
+    lmstudio_url = _lmstudio_server_root(normalized)
 
     cached = _endpoint_probe_path_cache.get(server_url)
     if cached is not None:

tests/agent/test_model_metadata_local_ctx.py

@@ -507,6 +507,66 @@ class TestDetectLocalServerTypeAuth:
         assert client_mock.get.call_args_list[0].args[0] == "http://localhost:1234/api/v1/models"
 
 
+class TestDetectLocalServerTypeLocalhostIPv4:
+    """detect_local_server_type should resolve localhost to 127.0.0.1."""
+
+    def test_localhost_resolved_to_ipv4(self):
+        """Probes should use 127.0.0.1, not localhost, to avoid IPv6 timeout."""
+        from agent.model_metadata import detect_local_server_type
+
+        resp = MagicMock()
+        resp.status_code = 200
+
+        client_mock = MagicMock()
+        client_mock.__enter__ = lambda s: client_mock
+        client_mock.__exit__ = MagicMock(return_value=False)
+        client_mock.get.return_value = resp
+
+        with patch("httpx.Client", return_value=client_mock):
+            detect_local_server_type("http://localhost:8317/v1")
+
+        for call in client_mock.get.call_args_list:
+            url = call[0][0]
+            assert "localhost" not in url, f"Probe URL still uses localhost: {url}"
+            assert "127.0.0.1" in url
+
+    def test_non_localhost_urls_unchanged(self):
+        """Non-localhost URLs should not be modified."""
+        from agent.model_metadata import detect_local_server_type
+
+        client_mock = MagicMock()
+        client_mock.__enter__ = lambda s: client_mock
+        client_mock.__exit__ = MagicMock(return_value=False)
+        resp = MagicMock()
+        resp.status_code = 404
+        client_mock.get.return_value = resp
+
+        with patch("httpx.Client", return_value=client_mock):
+            detect_local_server_type("http://192.168.1.100:8080")
+
+        for call in client_mock.get.call_args_list:
+            url = call[0][0]
+            assert "192.168.1.100" in url
+
+    def test_127_0_0_1_urls_unchanged(self):
+        """URLs already using 127.0.0.1 should pass through unchanged."""
+        from agent.model_metadata import detect_local_server_type
+
+        client_mock = MagicMock()
+        client_mock.__enter__ = lambda s: client_mock
+        client_mock.__exit__ = MagicMock(return_value=False)
+        resp = MagicMock()
+        resp.status_code = 404
+        client_mock.get.return_value = resp
+
+        with patch("httpx.Client", return_value=client_mock):
+            detect_local_server_type("http://127.0.0.1:8317")
+
+        for call in client_mock.get.call_args_list:
+            url = call[0][0]
+            assert "127.0.0.1" in url
+
+
 class TestFetchEndpointModelMetadataLmStudio:
     """fetch_endpoint_model_metadata should use LM Studio's native models endpoint."""