fix(provider): restore parameter transparency in core LLM provider adapters

核心对话适配器（OpenAI, Anthropic, Gemini）在准备请求 Payload 时未对 kwargs 进行合并，导致插件层传入的自定义参数（如 max_tokens, temperature, timeout 等）失效，回退到提供商的保守默认值。本次修复确保了各主流模型适配器对请求参数的完整透传。

Author: SXP-Simon

astrbot/core/provider/sources/anthropic_source.py

@@ -516,6 +516,7 @@ async def text_chat(
         model = model or self.get_model()
 
         payloads = {"messages": new_messages, "model": model}
+        payloads.update(kwargs)
 
         # Anthropic has a different way of handling system prompts
         if system_prompt:
@@ -572,6 +573,7 @@ async def text_chat_stream(
         model = model or self.get_model()
 
         payloads = {"messages": new_messages, "model": model}
+        payloads.update(kwargs)
 
         # Anthropic has a different way of handling system prompts
         if system_prompt:

astrbot/core/provider/sources/gemini_source.py

@@ -758,6 +758,7 @@ async def text_chat(
         model = model or self.get_model()
 
         payloads = {"messages": context_query, "model": model}
+        payloads.update(kwargs)
 
         retry = 10
         keys = self.api_keys.copy()
@@ -813,6 +814,7 @@ async def text_chat_stream(
         model = model or self.get_model()
 
         payloads = {"messages": context_query, "model": model}
+        payloads.update(kwargs)
 
         retry = 10
         keys = self.api_keys.copy()

astrbot/core/provider/sources/openai_source.py

@@ -619,8 +619,8 @@ async def _prepare_chat_payload(
                     context_query.extend(tcr.to_openai_messages())
 
         model = model or self.get_model()
-
         payloads = {"messages": context_query, "model": model}
+        payloads.update(kwargs)
 
         self._finally_convert_payload(payloads)