fix: route minimax m3 reasoning controls through profile

Follow up PR #46609's api.minimax.io reasoning report by moving the behavior out of the broad run_agent host gate and into the MiniMax provider profile. Only MiniMax-M3 on the documented OpenAI-compatible /v1 route gets reasoning_split/thinking/reasoning_effort; Anthropic-format MiniMax and non-M3 models keep their existing wire shapes.

Co-authored-by: goku94123 <gooku94123@gmail.com>

run_agent.py

@@ -4792,15 +4792,6 @@ class AIAgent:
                 return bool(github_model_reasoning_efforts(self.model))
             except Exception:
                 return False
-        if base_url_host_matches(self._base_url_lower, "api.minimax.io"):
-            # MiniMax (api.minimax.io): enable reasoning extra_body
-            # (reasoning_split, thinking, reasoning_effort) for both the
-            # Anthropic-format and OpenAI-format endpoints. Without this the
-            # safety gate strips those fields before they reach the API, so M3
-            # leaks thinking into response content and burns output tokens. M3
-            # specifically benefits from the OpenAI-compatible endpoint
-            # (/v1/chat/completions) for prompt caching.
-            return True
         if (self.provider or "").strip().lower() == "lmstudio":
             opts = self._lmstudio_reasoning_options_cached()
             # "off-only" (or absent) means no real reasoning capability.