fix: put the tier back where "on" by itself is not on

The minimax wire lost its tier value on the assumption that these models
reason by default, so injecting nothing still reads as "on". T10 measured
the real gateway and the assumption does not hold: MiniMax-M3 with no
reasoning parameter did not reason in 5 of 5 rounds, while all six
strength values worked. Existing downstreams on ENABLE_THINKING=true
went from reasoning to silently not reasoning, and the capability table
cannot catch it because phase 5 lets auto through unconditionally.

Restore on_base to the old {"reasoning_effort": "medium"} verbatim. This
is a stopgap - it hands the tier choice back to the library, which this
work set out to remove. The real fix is to constrain auto by the
capability table, a public behaviour change tracked as issue #21.

The assertions that said "minimax injects no tier on the on-tier" go
back with it; each carries a note on why it moved twice.

tests/unit/test_providers.py

@@ -46,13 +46,28 @@ class TestDefaultProfiles:
         assert w.effort_key == "reasoning_effort"
 
     def test_openai_family_sends_the_standard_field_only(self):
-        """gpt/claude/gemini 经网关都吃 OpenAI 标准的 reasoning_effort,不下发厂商方言。"""
-        for name in ("openai", "anthropic", "google", "minimax"):
+        """gpt/claude/gemini 经网关都吃 OpenAI 标准的 reasoning_effort,不下发厂商方言。
+
+        minimax 2026-09-05 起不在本组: 它的形态相同,但"开"这一档被迫带上了一个
+        档位值(见 `test_minimax_on_tier_carries_a_tier_value`)。
+        """
+        for name in ("openai", "anthropic", "google"):
             w = get_provider(name).thinking
             assert w.on_base == {}, name
             assert w.off == {"reasoning_effort": "none"}, name
             assert w.effort_key == "reasoning_effort", name
 
+    def test_minimax_on_tier_carries_a_tier_value(self):
+        """issue #21 的权宜之计: minimax 的"开"必须真写一个档位值,不能是空片段。
+
+        断言反复过一次: T2 按"这些模型默认就推理"的推定把它改成 `{}`,T10 真实
+        网关实测推翻推定(M3 不发推理参数时 5/5 轮不推理),故逐字恢复旧版的 medium。
+        """
+        w = get_provider("minimax").thinking
+        assert w.on_base == {"reasoning_effort": "medium"}
+        assert w.off == {"reasoning_effort": "none"}
+        assert w.effort_key == "reasoning_effort"
+
     def test_unknown_provider_fails_loudly(self):
         """消灭子串猜测: 未注册 provider 装配期即报错,不做模糊匹配。"""
         with pytest.raises(ValueError, match="glm"):
@@ -82,8 +97,12 @@ class TestWireNoneSemantics:
         assert get_provider("qwen").thinking.effort_key is None
 
     def test_empty_on_base_is_not_none(self):
-        """`{}` = 已知无需注入任何参数即处于该档;`None` = 不知道怎么表达。"""
-        w = get_provider("minimax").thinking
+        """`{}` = 已知无需注入任何参数即处于该档;`None` = 不知道怎么表达。
+
+        样本 2026-09-05 由 minimax 换成 openai: minimax 的 `on_base` 因 issue #21
+        改回带值,不再是空片段;openai 段是现存 `{}` 语义的代表。
+        """
+        w = get_provider("openai").thinking
         assert w.on_base == {} and w.on_base is not None