[Bug] Model change ignored when resubmitting an edited message after Stop

提交前确认 · Pre-submission checklist

• 我已搜索过现有 issue，确认这不是重复 / I searched existing issues and confirmed this isn't a duplicate.
• 我已阅读 CONTRIBUTING.md / I've read CONTRIBUTING.md.

问题类别 · Category

对话 / Agent 交互 · Agent chat

涉及的 Agent 框架 · Agent framework

ZCode Agent（自研）

严重程度 · Severity

阻塞使用 · Blocking (无法使用核心功能 / core function unusable)

复现频率 · Reproducibility

必现 · Always

问题描述 · Description

When a message is stopped mid-generation and then edited before being resubmitted, any model change made at that point is not applied — the resubmitted message still runs on the previously selected model instead of the newly selected one.

Describe the bug

Changing the model after stopping and editing a message does not update the model actually used for the resubmitted request.

Impact

Users cannot reliably switch models mid-conversation via stop-and-edit, which may cause unexpected model usage, inconsistent outputs, or unintended cost/quota impact if a different (e.g. pricier) model was intended.

复现步骤 · Steps to reproduce

1. Send a message and let generation begin.
2. Click Stop to halt generation.
3. Edit the message.
4. Change the selected model.
5. Submit the edited message.
6. Observe that the response is generated using the old model, not the newly selected one.

期望表现 · Expected behavior

After stopping a generation, editing the message, and selecting a different model, resubmitting should send the request using the newly selected model.

实际表现 · Actual behavior

The resubmitted message continues to use the model that was active before the stop/edit, ignoring the newly selected model.

ZCode 版本 · ZCode version

v3.12.3

设备 / 系统 / 浏览器 · Device / OS / Browser

Windows 11 25H2

截图 / 录屏 / 日志 · Screenshots / Recordings / Logs

No response

Start by reproducing the stop, edit, model-change, and resubmit sequence described in the issue, then trace how the selected model is carried into the resubmitted request. Confirm the fix by repeating the scenario and verifying that the request uses the newly selected model rather than the model active before stopping.