formatter: skip temperature override for reasoning models

GPT-5 and the o-series reject temperature=0 with a 400 BadRequestError —
only the default (1) is supported for reasoning models. Add an
_is_reasoning_model helper and pass temperature only when the model name
does not start with gpt-5/o1/o3/o4. Determinism on reasoning models is a
property of the architecture, not a parameter.

Discovered on the first real LLM run against the seed exemplars.

src/cmos/formatter.py

@@ -88,6 +88,17 @@ def build_user_message(messy_entry: str) -> str:
     )
 
 
+def _is_reasoning_model(model: str) -> bool:
+    """Reasoning-model families (GPT-5, o-series) don't accept temperature overrides.
+
+    OpenAI rejects ``temperature=0`` on these with a 400 BadRequestError;
+    only the default (1) is supported. Determinism on reasoning models is a
+    property of the architecture, not a parameter.
+    """
+    prefixes = ("gpt-5", "o1", "o3", "o4")
+    return any(model.startswith(p) for p in prefixes)
+
+
 def _openai_caller(system: str, user: str) -> str:
     """Default caller — hits the real OpenAI API.
 
@@ -99,14 +110,18 @@ def _openai_caller(system: str, user: str) -> str:
     from openai import OpenAI  # imported lazily so unit tests do not need the network
 
     client = OpenAI()
-    response = client.chat.completions.create(
-        model=MODEL,
-        temperature=0,
-        messages=[
+    kwargs: dict = {
+        "model": MODEL,
+        "messages": [
             {"role": "system", "content": system},
             {"role": "user", "content": user},
         ],
-    )
+    }
+    # Non-reasoning models still benefit from temperature=0; reasoning models
+    # reject it.
+    if not _is_reasoning_model(MODEL):
+        kwargs["temperature"] = 0
+    response = client.chat.completions.create(**kwargs)
     return response.choices[0].message.content or ""