feat: 新增 Codex 上游兼容整流器

- 强制 store=false
- 确保 instructions 字段存在（缺失/None -> ""）
- 强制 parallel_tool_calls=true，并确保 include 包含 reasoning.encrypted_content
- 删除 Codex 会拒绝的一些字段：max_output_tokens/max_completion_tokens/max_tokens/temperature/top_p/service_tier
- 将 input[] 里 role=system 的 message 改成 developer
- 在发送上游请求前注入整流（覆盖“同格式透传”和“跨格式转换”两条链路）

tests/services/test_codex_request.py

@@ -0,0 +1,103 @@
+from __future__ import annotations
+
+from src.services.provider.codex import maybe_patch_request_for_codex, patch_openai_cli_request_for_codex
+
+
+def test_patch_openai_cli_request_for_codex_sets_store_and_instructions() -> None:
+    req = {"model": "gpt-test", "input": []}
+    out = patch_openai_cli_request_for_codex(req)
+
+    assert out is not req
+    assert out["store"] is False
+    assert "instructions" in out
+    assert isinstance(out["instructions"], str)
+
+
+def test_patch_openai_cli_request_for_codex_strips_rejected_params() -> None:
+    req = {
+        "model": "gpt-test",
+        "input": [],
+        "max_output_tokens": 123,
+        "max_completion_tokens": 456,
+        "max_tokens": 789,
+        "temperature": 0.5,
+        "top_p": 0.9,
+        "service_tier": "default",
+    }
+    out = patch_openai_cli_request_for_codex(req)
+
+    for key in (
+        "max_output_tokens",
+        "max_completion_tokens",
+        "max_tokens",
+        "temperature",
+        "top_p",
+        "service_tier",
+    ):
+        assert key not in out
+
+
+def test_patch_openai_cli_request_for_codex_converts_system_role_to_developer() -> None:
+    req = {
+        "model": "gpt-test",
+        "instructions": "ignored",
+        "input": [
+            {
+                "type": "message",
+                "role": "system",
+                "content": [{"type": "input_text", "text": "You are a pirate."}],
+            },
+            {
+                "type": "message",
+                "role": "user",
+                "content": [{"type": "input_text", "text": "Hello"}],
+            },
+        ],
+    }
+    out = patch_openai_cli_request_for_codex(req)
+
+    assert isinstance(out.get("input"), list)
+    assert out["input"][0]["role"] == "developer"
+    assert out["input"][1]["role"] == "user"
+
+
+def test_patch_openai_cli_request_for_codex_adds_required_include_item() -> None:
+    req = {"model": "gpt-test", "input": []}
+    out = patch_openai_cli_request_for_codex(req)
+
+    assert "include" in out
+    assert "reasoning.encrypted_content" in out["include"]
+
+
+def test_maybe_patch_request_for_codex_is_noop_for_non_codex() -> None:
+    req = {"model": "gpt-test", "input": []}
+    out = maybe_patch_request_for_codex(
+        provider_type="custom",
+        provider_api_format="openai:cli",
+        request_body=req,
+    )
+    assert out is req
+
+
+def test_maybe_patch_request_for_codex_is_noop_for_non_openai_cli() -> None:
+    req = {"model": "gpt-test", "input": []}
+    out = maybe_patch_request_for_codex(
+        provider_type="codex",
+        provider_api_format="openai:chat",
+        request_body=req,
+    )
+    assert out is req
+
+
+def test_maybe_patch_request_for_codex_patches_for_codex_openai_cli() -> None:
+    req = {"model": "gpt-test", "input": []}
+    out = maybe_patch_request_for_codex(
+        provider_type="codex",
+        provider_api_format="openai:cli",
+        request_body=req,
+    )
+
+    assert out is not req
+    assert out["store"] is False
+    assert "instructions" in out
+