Fix model connection credential reveal and save flow

[javaht]

修改内容

• 修复已保存模型密钥无法通过眼睛按钮查看的问题

  • 默认仍保持隐藏
  • 仅在用户主动点击显示按钮时读取当前 API key
  • OAuth token 不开放读取，仍保留在主进程

• 调整模型连接高级设置的保存逻辑

  • 模型勾选不再立即自动保存
  • 将“保存服务地址”改为“保存”
  • 点击保存时一次提交启用模型列表和服务地址

• 改进保存后的交互反馈

  • 保存成功后自动关闭连接详情页
  • 右下角显示“连接名称 已保存”提示
  • 保存或刷新失败时保留页面和未提交内容，方便继续修改

• 补充配套改动

  • 更新中英文文案和安全边界说明
  • 更新 Storybook mock
  • 增加 API key 读取边界、统一保存及成功反馈的回归测试

图一图二是前后对比

[Astro-Han]

The credential reveal is blocked by the inline security finding.

[P2] The new tests inspect source strings instead of exercising IPC and renderer behavior. connection-credential-ipc-hardening-contract.test.ts explicitly requires credentialStore.getSecret() in a renderer-facing handler, so the test locks in the boundary violation rather than catching it. After removing the reveal path, add behavior-level coverage that invokes the registered IPC handler and verifies that no main-to-renderer result contains cleartext. The reveal and save states also need a component or E2E test instead of source-shape assertions.

Please revise before merging.

[Astro-Han]

[P1] Keep API keys on the main side of IPC

This returns the stored API key from main to the renderer. SECURITY.md sections 2.3 and 4 treat the renderer as semi-trusted and require cleartext secrets to travel only from renderer to main. A compromised renderer can list connection slugs, call this handler for each API-key provider, and read the saved credentials. Keep the stored value in main, expose only masked or presence state, and accept a replacement value from the renderer when the user saves.