refactor: centralize llm backend configuration

plugins/ai_auto_response/config.toml

@@ -9,17 +9,7 @@ max_reply_sentences = 3
 familiarity_hint = "有熟悉感，但不过度装熟"
 
 [api]
-provider = "openai_compatible"
-api_base_url = "http://192.168.2.240:3000/v1"
-endpoint = "chat/completions"
-api_key = "sk-hC6WMLAsTdItpywyrYdxT6pQ4E7NARGbUKuPWRH0zMheen9e"
-model = "gpt-5.4"
-timeout_seconds = 45
-temperature = 0.35
-max_tokens = 120
-stream = true
-max_retries = 3
-retry_delay_seconds = 1.0
+backend = "openai_compatible_ai_auto_response"
 
 [mode]
 group_default_mode = "social"