fix(llm): 删除未实现的 LlmRequest.thinking 字段（CR-H5，YAGNI）

ARCHITECTURE.md

@@ -481,7 +481,6 @@ class LlmRequest(BaseModel):
     input: str | list[Block]                 # 易变内容(断点之后)
     stream: bool = False
     output_schema: type[BaseModel] | None = None  # 需结构化输出时(Pydantic 模型/instructor)
-    thinking: bool = False                   # 是否启用思考/推理(有则启用)
     max_tokens: int | None = None
     scope: "Scope"                           # {project_id; user_id 原型可固定}
 

memory/contracts.md

@@ -10,7 +10,7 @@
 - 消费方：编排器、所有 Agent（经 `Gateway.run` / `Gateway.stream`）。
 - 关键不变量：agent 只传 `tier`（writer/analyst/light），不传具体 model。
 - **已实现（`packages/llm_gateway/ww_llm_gateway`）**：
-  - 类型 `types.py`：`Block(text,cache=False)`、`Scope(user_id,project_id?)`、`LlmRequest(tier,input:str|list[Block],system:list[Block],stream,output_schema?,thinking,max_tokens?,scope)`、`Usage(provider,model,input_tokens,output_tokens,cache_read_tokens,cost_minor,currency)`、`ServedBy(provider,model,fell_back)`、`LlmResponse(text,parsed?,usage,served_by)`、`Delta(text)`、`Tier`。
+  - 类型 `types.py`：`Block(text,cache=False)`、`Scope(user_id,project_id?)`、`LlmRequest(tier,input:str|list[Block],system:list[Block],stream,output_schema?,max_tokens?,scope)`、`Usage(provider,model,input_tokens,output_tokens,cache_read_tokens,cost_minor,currency)`、`ServedBy(provider,model,fell_back)`、`LlmResponse(text,parsed?,usage,served_by)`、`Delta(text)`、`Tier`。
   - `Gateway(adapters:dict[str,ProviderAdapter], ledger:LedgerSink, resolver=resolve_route)`：`async run(req)->LlmResponse`、`stream(req)->AsyncIterator[Delta]`。每次调用落 **1 条** `usage_ledger`（经 `LedgerSink`，可注入内存替身）。
   - 适配器 `ProviderAdapter`(Protocol)：`provider`、`capabilities()->Capabilities`、`async complete(req,model)->ProviderResult`、`stream(req,model)->AsyncIterator[StreamChunk]`。M1 实现 `OpenAICompatAdapter(provider, client:AsyncOpenAI)`（DeepSeek，注入 client 便于测试）。
   - 档位路由 `resolve_route(tier)->Route(provider,model)` 读 `config.tier_defaults`（M1 仅全局默认；回退/熔断属 M5/T5.4，**未实现**）。

packages/llm_gateway/tests/test_types.py

@@ -0,0 +1,22 @@
+"""C1 类型契约不变量（`LlmRequest` 字段集）单测。"""
+
+from __future__ import annotations
+
+from ww_llm_gateway.adapters.base import Capabilities
+from ww_llm_gateway.types import LlmRequest
+
+
+def test_llm_request_has_no_thinking_field() -> None:
+    # CR-H5：删除从未被任何适配器/网关读取的死字段 `LlmRequest.thinking`（YAGNI）。
+    assert "thinking" not in LlmRequest.model_fields
+    assert set(LlmRequest.model_fields) == {
+        "tier",
+        "input",
+        "system",
+        "stream",
+        "output_schema",
+        "max_tokens",
+        "scope",
+    }
+    # 同名但另一处 live 字段 `Capabilities.thinking`（适配器能力矩阵）不受影响。
+    assert "thinking" in Capabilities.model_fields

packages/llm_gateway/ww_llm_gateway/types.py

@@ -38,7 +38,6 @@ class LlmRequest(BaseModel):
     system: list[Block] = Field(default_factory=list)
     stream: bool = False
     output_schema: type[BaseModel] | None = None
-    thinking: bool = False
     max_tokens: int | None = None
     scope: Scope