Restore OpenAI params to focus PR on tool-call fix

Author: latent-variable

mini_agent/llm/openai_client.py

@@ -65,12 +65,8 @@ async def _make_api_request(
         params = {
             "model": self.model,
             "messages": api_messages,
-            # Add consistent parameters for better KV cache stability
-            "temperature": 0.7,  # Consistent temperature
-            "seed": 42,  # Fixed seed for deterministic behavior and better caching
             # Enable reasoning_split to separate thinking content
-            # NOTE: Commenting out for local models - may cause cache invalidation
-            # "extra_body": {"reasoning_split": True},
+            "extra_body": {"reasoning_split": True},
         }
 
         if tools: