docs: retire the criterion this version disproved, everywhere it survived

The reasoning_tokens docstring was still teaching downstream to treat
None or 0 as no reasoning. The changelog and the schema page had both
been corrected; the docstring had not, and it is the copy that ships in
the wheel and shows up on hover. Someone writing a report from it would
have counted every real MiniMax reasoning call as not reasoning, which
is issue #16 all over again with the tests green.

The original wording stays, since reading pre-1.3.1 rows still needs
it. What follows it now says when it expired and what to read instead.

Two more places had drifted the same way: the changelog and the
architecture doc described the throttle and the cache fallback as they
were before this review, which is to say as the opposite of what the
code now does.

The claim that the two throttle sets would suppress each other does not
survive checking, as the mutation testing showed: their key spaces do
not overlap. Keeping them apart is still right, but for the honest
reason, which is that the two warnings have unrelated lifetimes.

research-wiki/ARCHITECTURE.md

@@ -393,9 +393,9 @@ flowchart TB
 
 三态**不可折叠为布尔**: `unknown`(判不出)与 `absent`(确证没有)语义不同,把前者读作后者正是 `reasoning_tokens=None` 制造的那个歧义——MiniMax-M3 非流式开启推理时,推理内容已计费却不回传正文(2026-08-25 实测 completion 53 vs 关闭档 3),该档只能判 `unknown`,宣称「没推理」即撒谎。缺省取 `UNKNOWN` 使任何不填该字段的路径(非 OpenAI 兼容 transport、失败尝试、终态失败行)天然诚实——**默认值本身不撒谎**,这是 P5 在字段设计上的落法。
 
-判据取 `thinking.strip()` 而非 `bool(thinking)`: transport 收集 `reasoning_content` 时只判 truthy,上游返回纯空白串会被计成「观测到推理」(网关响应是外部输入,校验后使用)。裁定纯函数 `observe_thinking` 定义在 `thinking.py`,由 `openai_compat` 的流式与非流式**两条**组装路径各调一次(只填一条即分叉);`CacheMW._rehydrate` 回放时显式转回枚举实例(JSON 复活的是裸 `str`),非法值转换失败由既有 try/except 吞成「按未命中回源」。该字段**不进缓存 key**——它是结果不是请求。
+判据取 `thinking.strip()` 而非 `bool(thinking)`: transport 收集 `reasoning_content` 时只判 truthy,上游返回纯空白串会被计成「观测到推理」(网关响应是外部输入,校验后使用)。裁定纯函数 `observe_thinking` 定义在 `thinking.py`,由 `openai_compat` 的流式与非流式**两条**组装路径各调一次(只填一条即分叉);`CacheMW._rehydrate` 回放时显式转回枚举实例(JSON 复活的是裸 `str`),域外取值降级为 `unknown` 并单独告警、内容照常复活——纯可观测性字段不该有能力作废内容完好的缓存(多项目共用同一 Redis 时,先升级者写入的新态会让未升级者每次判未命中、覆写回旧值,两版互打缓存);「整条作废」只留给真正破坏内容完整性的失败。该字段**不进缓存 key**——它是结果不是请求。
 
-**声明 × 观测对账(同批)**: `reconcile_thinking` 把请求方向(`enable_thinking`)与实测观测比对,矛盾即 warning、**不抛错**(可观测性属遥测方向,降级即 warning;且一次观测不足以否决一次成功的调用)。四种矛盾各有独立文案: 关闭请求却观测到推理(已登记 / 未登记两说,后者不得声称「能力表声称可关闭」——它根本没登记)、开启却上报未推理、开启却观测不到。`False × unknown` 与 `None × 任意` **不表态**: `unknown` 没有证伪力,拿它报警等于每次关闭调用都喊一遍,噪声即等于没有告警。节流按 per-transport-instance 的 `(model, direction)` 集合,与既有 `_warned_models` 同款形态但**不可复用同一个集合**(两者语义不同,混用会互相压制)。
+**声明 × 观测对账(同批)**: `reconcile_thinking` 把请求方向(`enable_thinking`)与实测观测比对,矛盾即 warning、**不抛错**(可观测性属遥测方向,降级即 warning;且一次观测不足以否决一次成功的调用)。四种矛盾各有独立文案: 关闭请求却观测到推理(已登记 / 未登记两说,后者不得声称「能力表声称可关闭」——它根本没登记)、开启却上报未推理、开启却观测不到。`False × unknown` 与 `None × 任意` **不表态**: `unknown` 没有证伪力,拿它报警等于每次关闭调用都喊一遍,噪声即等于没有告警。节流按 per-transport-instance 的 `(source, model, direction)` 集合,与既有 `_warned_models` 同款形态但**不可复用同一个集合**(两者语义不同——一个记「未登记能力已告警过」,一个记「某源某方向的矛盾已告警过」,共用会让两种告警的生命周期纠缠;键空间本就不相交,故不是碰撞问题)。键含源名是因为多源多账号是本库的核心场景: 同一 model 跨 N 个源常态,漏掉源名会让第一个出问题的源喊完之后其余源永久静音,且告警定位不到该查哪个网关(源名在调用点拼进文案,不进纯判定函数的签名)。
 
 这条对账的价值在于把「能力表过期」从**静默错觉**变成日志里的显式告警——能力表过期是必然事件(M3 的 evidence 曾停在 8-02 整整 23 天),成本是一次枚举比较。但**保障只覆盖可观测路径**: M3 非流式两个信号双缺,那里的推理开关哪天失效库同样看不见,这一点不得假装有。