feat: add ResponsesApiTransport + wire all Codex transport paths

Add ResponsesApiTransport wrapping codex_responses_adapter.py behind the
ProviderTransport ABC. Auto-registered via _discover_transports().

Wire ALL Codex transport methods to production paths in run_agent.py:
- build_kwargs: main _build_api_kwargs codex branch (50 lines extracted)
- normalize_response: main loop + flush + summary + retry (4 sites)
- convert_tools: memory flush tool override
- convert_messages: called internally via build_kwargs
- validate_response: response validation gate
- preflight_kwargs: request sanitization (2 sites)

Remove 7 dead legacy wrappers from AIAgent (_responses_tools,
_chat_messages_to_responses_input, _normalize_codex_response,
_preflight_codex_api_kwargs, _preflight_codex_input_items,
_extract_responses_message_text, _extract_responses_reasoning_text).
Keep 3 ID manipulation methods still used by _build_assistant_message.

Update 18 test call sites across 3 test files to call adapter functions
directly instead of through deleted AIAgent wrappers.

24 new tests. 343 codex/responses/transport tests pass (0 failures).

PR 4 of the provider transport refactor.

tests/run_agent/test_run_agent_codex_responses.py

@@ -640,7 +640,8 @@ def test_run_conversation_codex_tool_round_trip(monkeypatch):
 
 def test_chat_messages_to_responses_input_uses_call_id_for_function_call(monkeypatch):
     agent = _build_agent(monkeypatch)
-    items = agent._chat_messages_to_responses_input(
+    from agent.codex_responses_adapter import _chat_messages_to_responses_input
+    items = _chat_messages_to_responses_input(
         [
             {"role": "user", "content": "Run terminal"},
             {
@@ -668,7 +669,8 @@ def test_chat_messages_to_responses_input_uses_call_id_for_function_call(monkeyp
 
 def test_chat_messages_to_responses_input_accepts_call_pipe_fc_ids(monkeypatch):
     agent = _build_agent(monkeypatch)
-    items = agent._chat_messages_to_responses_input(
+    from agent.codex_responses_adapter import _chat_messages_to_responses_input
+    items = _chat_messages_to_responses_input(
         [
             {"role": "user", "content": "Run terminal"},
             {
@@ -696,7 +698,8 @@ def test_chat_messages_to_responses_input_accepts_call_pipe_fc_ids(monkeypatch):
 
 def test_preflight_codex_api_kwargs_strips_optional_function_call_id(monkeypatch):
     agent = _build_agent(monkeypatch)
-    preflight = agent._preflight_codex_api_kwargs(
+    from agent.codex_responses_adapter import _preflight_codex_api_kwargs
+    preflight = _preflight_codex_api_kwargs(
         {
             "model": "gpt-5-codex",
             "instructions": "You are Hermes.",
@@ -724,7 +727,8 @@ def test_preflight_codex_api_kwargs_rejects_function_call_output_without_call_id
     agent = _build_agent(monkeypatch)
 
     with pytest.raises(ValueError, match="function_call_output is missing call_id"):
-        agent._preflight_codex_api_kwargs(
+        from agent.codex_responses_adapter import _preflight_codex_api_kwargs
+        _preflight_codex_api_kwargs(
             {
                 "model": "gpt-5-codex",
                 "instructions": "You are Hermes.",
@@ -741,7 +745,8 @@ def test_preflight_codex_api_kwargs_rejects_unsupported_request_fields(monkeypat
     kwargs["some_unknown_field"] = "value"
 
     with pytest.raises(ValueError, match="unsupported field"):
-        agent._preflight_codex_api_kwargs(kwargs)
+        from agent.codex_responses_adapter import _preflight_codex_api_kwargs
+        _preflight_codex_api_kwargs(kwargs)
 
 
 def test_preflight_codex_api_kwargs_allows_reasoning_and_temperature(monkeypatch):
@@ -752,7 +757,8 @@ def test_preflight_codex_api_kwargs_allows_reasoning_and_temperature(monkeypatch
     kwargs["temperature"] = 0.7
     kwargs["max_output_tokens"] = 4096
 
-    result = agent._preflight_codex_api_kwargs(kwargs)
+    from agent.codex_responses_adapter import _preflight_codex_api_kwargs
+    result = _preflight_codex_api_kwargs(kwargs)
     assert result["reasoning"] == {"effort": "high", "summary": "auto"}
     assert result["include"] == ["reasoning.encrypted_content"]
     assert result["temperature"] == 0.7
@@ -764,7 +770,8 @@ def test_preflight_codex_api_kwargs_allows_service_tier(monkeypatch):
     kwargs = _codex_request_kwargs()
     kwargs["service_tier"] = "priority"
 
-    result = agent._preflight_codex_api_kwargs(kwargs)
+    from agent.codex_responses_adapter import _preflight_codex_api_kwargs
+    result = _preflight_codex_api_kwargs(kwargs)
     assert result["service_tier"] == "priority"
 
 
@@ -841,7 +848,8 @@ def test_run_conversation_codex_continues_after_incomplete_interim_message(monke
 
 def test_normalize_codex_response_marks_commentary_only_message_as_incomplete(monkeypatch):
     agent = _build_agent(monkeypatch)
-    assistant_message, finish_reason = agent._normalize_codex_response(
+    from agent.codex_responses_adapter import _normalize_codex_response
+    assistant_message, finish_reason = _normalize_codex_response(
         _codex_commentary_message_response("I'll inspect the repository first.")
     )
 
@@ -1068,7 +1076,8 @@ def test_normalize_codex_response_marks_reasoning_only_as_incomplete(monkeypatch
     sends them into the empty-content retry loop (3 retries then failure).
     """
     agent = _build_agent(monkeypatch)
-    assistant_message, finish_reason = agent._normalize_codex_response(
+    from agent.codex_responses_adapter import _normalize_codex_response
+    assistant_message, finish_reason = _normalize_codex_response(
         _codex_reasoning_only_response()
     )
 
@@ -1101,7 +1110,8 @@ def test_normalize_codex_response_reasoning_with_content_is_stop(monkeypatch):
         status="completed",
         model="gpt-5-codex",
     )
-    assistant_message, finish_reason = agent._normalize_codex_response(response)
+    from agent.codex_responses_adapter import _normalize_codex_response
+    assistant_message, finish_reason = _normalize_codex_response(response)
 
     assert finish_reason == "stop"
     assert "Here is the answer" in assistant_message.content
@@ -1186,7 +1196,8 @@ def test_chat_messages_to_responses_input_reasoning_only_has_following_item(monk
             ],
         },
     ]
-    items = agent._chat_messages_to_responses_input(messages)
+    from agent.codex_responses_adapter import _chat_messages_to_responses_input
+    items = _chat_messages_to_responses_input(messages)
 
     # Find the reasoning item
     reasoning_indices = [i for i, it in enumerate(items) if it.get("type") == "reasoning"]
@@ -1273,7 +1284,8 @@ def test_chat_messages_to_responses_input_deduplicates_reasoning_ids(monkeypatch
             ],
         },
     ]
-    items = agent._chat_messages_to_responses_input(messages)
+    from agent.codex_responses_adapter import _chat_messages_to_responses_input
+    items = _chat_messages_to_responses_input(messages)
 
     reasoning_items = [it for it in items if it.get("type") == "reasoning"]
     # Dedup: rs_aaa appears in both turns but should only be emitted once.
@@ -1299,7 +1311,8 @@ def test_preflight_codex_input_deduplicates_reasoning_ids(monkeypatch):
         {"type": "reasoning", "id": "rs_zzz", "encrypted_content": "enc_b"},
         {"role": "assistant", "content": "done"},
     ]
-    normalized = agent._preflight_codex_input_items(raw_input)
+    from agent.codex_responses_adapter import _preflight_codex_input_items
+    normalized = _preflight_codex_input_items(raw_input)
 
     reasoning_items = [it for it in normalized if it.get("type") == "reasoning"]
     # rs_xyz duplicate should be collapsed to one item; rs_zzz kept.