fix: normalise a bare tier string at the fourth door

`resolve_thinking` is exported, and its third argument turned from `bool`
into `Effort` this cycle — so the most natural downstream call now passes
the `"low"` it read out of JSON or config. Every gate inside compares by
identity, and `"none" is Effort.NONE` is always false: the tier was not
rejected, it was silently misjudged. Phase 2 read the on-form for a
request that wanted the off-form, Phase 4 was skipped entirely on a model
that cannot be disabled, and the only visible symptom arrived much later
as an AttributeError from `.value` — an exception that is neither
documented nor one of the four error classes.

Design 4.4 already lists this as the fourth entry a tier can come in
through; this makes the code agree with it.

src/polygateway/thinking.py

@@ -15,7 +15,7 @@ from typing import Any
 from loguru import logger
 
 from polygateway.providers import ProviderProfile, ThinkingWire
-from polygateway.types import EFFORT_ORDER, Effort, ThinkingObservation
+from polygateway.types import EFFORT_ORDER, Effort, ThinkingObservation, coerce_effort
 
 
 def observe_thinking(*, thinking: str, reasoning_tokens: int | None) -> ThinkingObservation:
@@ -344,7 +344,7 @@ def effective_effort(
 def resolve_thinking(
     profile: ProviderProfile,
     capability: ThinkingCapability | None,
-    effort: Effort | None,
+    effort: Effort | str | None,
     *,
     model: str,
     fallback: str = "error",
@@ -385,7 +385,17 @@ def resolve_thinking(
 
     `warn_unregistered=False` 供请求热路径去重用: 装配期已经喊过一次,逐次调用
     再喊只会刷屏。判定结果不受此参数影响。
+
+    **裸字符串也收**(设计 §4.4 第 4 条入口): 本函数在 `__all__` 里,下游直调时
+    传的天然是从 JSON/配置读出来的 `"low"`,而第三参数本次由 `bool` 换成 `Effort`
+    正是这条入口冒出来的时机。签名照实写 `Effort | str`——下面每一关的判据都是
+    `is Effort.X` 的身份比较,`"none" is Effort.NONE` 恒假,不归一的后果不是报错
+    而是**静默判否**: Phase 2 按开启方向取字段、Phase 4 整条被绕过,最后在拼错误
+    文案时才以 `AttributeError` 现形(一个未文档化、也不属四分类的异常)。
     """
+    # Phase 0: 归一 —— 判据全是身份比较,入口不归一则后面每一关都在拿裸串比枚举
+    if effort is not None:
+        effort = coerce_effort(effort, origin=f"resolve_thinking(model={model!r})")
     # Phase 1: 调用方不表态 —— 与 Effort.NONE 严格区分,用模型自己的默认档
     if effort is None:
         return ThinkingResolution({}, None)