feat(inference): add main() — runs all 3 task difficulties with 20-minute guard

inference.py

@@ -61,3 +61,18 @@ async def run_episode(task: str = "easy") -> dict:
         
         grade_resp = await http.post("/grade")
         return {"task": task, "turns": turn, "grade": grade_resp.json()}
+
+async def main():
+    import time
+    start = time.time()
+    for task in ["easy", "medium", "hard"]:
+        logger.info(f"Running {task}...")
+        try:
+            await run_episode(task)
+        except Exception as e:
+            logger.error(f"Failed {task}: {e}")
+    if time.time() - start > 1200:
+        logger.warning("Exceeded 20 mins!")
+
+if __name__ == "__main__":
+    asyncio.run(main())