fix(model_metadata): rewrite localhost->IPv4 for the remaining local probe sites

fetch_endpoint_model_metadata's generic (non-LM-Studio) /models fetch and
its llama.cpp /v1/props context-length follow-up built request URLs
straight from the unrewritten candidate, unlike every other local-probe
site. Both retained the multi-second dual-stack IPv6 connect penalty
that _localhost_to_ipv4() exists to skip (measured on macOS: localhost
32.9ms vs 127.0.0.1 0.1ms on a dead port; ~2s on Windows). normalized
stays the cache key so caching behavior is unchanged; only the outbound
request target is rewritten.

Re-derived from PR #61528 onto current main (original no longer applied
cleanly).

agent/model_metadata.py

@@ -1184,7 +1184,11 @@ def fetch_endpoint_model_metadata(
             last_error = exc
 
     for candidate in candidates:
-        url = candidate.rstrip("/") + "/models"
+        # normalized/candidates stay unrewritten (cache key stability); only
+        # the outbound request target is IPv4-resolved to skip the multi-second
+        # dual-stack IPv6 connect timeout (see _localhost_to_ipv4).
+        request_candidate = _localhost_to_ipv4(candidate)
+        url = request_candidate.rstrip("/") + "/models"
         response = None
         try:
             response = requests.get(
@@ -1230,7 +1234,7 @@ def fetch_endpoint_model_metadata(
             if is_llamacpp:
                 try:
                     # Try /v1/props first (current llama.cpp); fall back to /props for older builds
-                    base = candidate.rstrip("/").replace("/v1", "")
+                    base = request_candidate.rstrip("/").replace("/v1", "")
                     _verify = _resolve_requests_verify()
                     props_resp = requests.get(base + "/v1/props", headers=headers, timeout=5, verify=_verify)
                     if not props_resp.ok:

tests/agent/test_probe_cache_followups.py

@@ -186,6 +186,61 @@ class TestLocalhostIPv4SiblingSites:
 
         assert client.post.call_args[0][0].startswith("http://127.0.0.1:11434")
 
+    def test_fetch_endpoint_model_metadata_generic_probe_uses_ipv4(self):
+        """The generic (non-LM-Studio) /models fetch loop must also rewrite
+        localhost->127.0.0.1 before probing, like the LM Studio branch above."""
+        from agent import model_metadata
+        from agent.model_metadata import fetch_endpoint_model_metadata
+
+        model_metadata._endpoint_model_metadata_cache.clear()
+        model_metadata._endpoint_model_metadata_cache_time.clear()
+
+        resp = MagicMock()
+        resp.status_code = 200
+        resp.raise_for_status = MagicMock()
+        resp.json.return_value = {"data": []}
+
+        with patch("agent.model_metadata.detect_local_server_type", return_value=None), \
+             patch("agent.model_metadata.requests.get", return_value=resp) as mock_get:
+            fetch_endpoint_model_metadata("http://localhost:8000/v1")
+
+        assert mock_get.call_args[0][0].startswith("http://127.0.0.1:8000")
+
+    def test_fetch_endpoint_model_metadata_llamacpp_props_followup_uses_ipv4(self):
+        """The llama.cpp /props context-length follow-up must also rewrite
+        localhost->127.0.0.1 before probing, not just the initial /models call."""
+        from agent import model_metadata
+        from agent.model_metadata import fetch_endpoint_model_metadata
+
+        model_metadata._endpoint_model_metadata_cache.clear()
+        model_metadata._endpoint_model_metadata_cache_time.clear()
+
+        models_resp = MagicMock()
+        models_resp.status_code = 200
+        models_resp.raise_for_status = MagicMock()
+        models_resp.json.return_value = {
+            "data": [{"id": "llama-3-8b", "owned_by": "llamacpp"}],
+        }
+
+        props_resp = MagicMock()
+        props_resp.ok = True
+        props_resp.json.return_value = {
+            "default_generation_settings": {"n_ctx": 32768},
+            "model_alias": "llama-3-8b",
+        }
+
+        with patch("agent.model_metadata.detect_local_server_type", return_value=None), \
+             patch(
+                 "agent.model_metadata.requests.get",
+                 side_effect=[models_resp, props_resp],
+             ) as mock_get:
+            result = fetch_endpoint_model_metadata("http://localhost:8000/v1")
+
+        assert mock_get.call_count == 2
+        props_call_url = mock_get.call_args_list[1][0][0]
+        assert props_call_url.startswith("http://127.0.0.1:8000")
+        assert result["llama-3-8b"]["context_length"] == 32768
+
 
 
 class TestContextCacheKeyNormalization: