feat: 添加失败时保存本轮记录的配置项

[Blueteemo]

问题描述

目前 Agent 运行失败时（例如 EmptyModelOutputError），本轮记录通常不会写入正式会话历史。这在排查问题和长流程任务失败后希望从断点继续的场景下带来不便。

修复方案

添加 save_failed_agent_history 配置项（默认 false），启用后即使 Agent 本轮运行失败，也会将本轮记录保存到会话历史中，包括用户输入、工具调用记录和失败提示。

修改内容

1. internal.py：读取配置并传递给 _save_to_history 方法
2. _save_to_history 方法：添加 save_failed_history 参数控制空响应时是否保存
3. default.py：添加配置项定义和 UI 元数据

Related Issue

Fixes #7620

Summary by Sourcery

Add configurable support for saving the current agent run to conversation history even when the run fails.

New Features:

• Introduce a save_failed_agent_history configuration flag to control whether failed agent runs are persisted to conversation history.

Enhancements:

• Wire the save_failed_agent_history flag through the agent internal pipeline so that empty or failed model responses can still be saved when enabled.
• Extend chat provider template metadata to expose the new flag in the UI with appropriate description and conditions.

[sourcery-ai]

Hey - I've left some high level feedback:

• In _save_to_history, when save_failed_history is true and the response is otherwise empty, consider adding an explicit debug/info log (e.g., including run status or error reason) to make it clear that a failed/empty run is being intentionally persisted.
• The UI config for provider_settings.save_failed_agent_history is currently conditioned on agent_runner_type == 'local'; if other runner types also support this behavior, you may want to relax or extend this condition to avoid the option silently not appearing for them.

Prompt for AI Agents

Please address the comments from this code review:

## Overall Comments
- In `_save_to_history`, when `save_failed_history` is true and the response is otherwise empty, consider adding an explicit debug/info log (e.g., including run status or error reason) to make it clear that a failed/empty run is being intentionally persisted.
- The UI config for `provider_settings.save_failed_agent_history` is currently conditioned on `agent_runner_type == 'local'`; if other runner types also support this behavior, you may want to relax or extend this condition to avoid the option silently not appearing for them.

---

Sourcery is free for open source - if you like our reviews please consider sharing them ✨

• X
• Mastodon
• LinkedIn
• Facebook

_{Help me be more useful! Please click 👍 or 👎 on each comment and I'll use the feedback to improve your reviews.}

[gemini-code-assist]

Code Review

This pull request introduces a new configuration option, save_failed_agent_history, which allows the system to save session history even when an agent fails or returns an empty response. The changes include updates to the default configuration, schema definitions, and the internal agent processing logic. However, the current implementation has two significant issues: the new parameter is not passed to _save_to_history in non-Live mode, and existing early-return logic within _save_to_history prevents the function from reaching the new logic when the LLM response is null or the message role is not 'assistant'.

[Soulter]

是否会导致下一轮的 messages 格式有误？

[Blueteemo]

感谢您的review！您担心的问题确实存在。

当 save_failed_history=True 且 response 为空时，我们之前只是创建了一个空的 LLMResponse 但没有添加到 messages 中保存。

现已修改：在保存失败历史时，会在 messages 中添加一条带有失败标记的 assistant message [Agent run failed. History saved for debugging.]，确保下一轮 messages 格式正确。