fix: skip image parts for DeepSeek to avoid unknown variant "image_url"

[JoeGlenn1213]

• 背景问题：DeepSeek 文本接口不支持图片输入，包含图片消息会触发 unknown variant "image_url"
• 修复内容：在构造用户消息时，检测当前 modelId ；若是 DeepSeek，不注入 type: "image" 内容
• 变更位置：服务器路由 app/api/chat/route.ts ，仅影响 DeepSeek，其他视觉模型不受影响
• 验证步骤：
• 配置 AI_PROVIDER=deepseek + AI_MODEL=deepseek-chat ，不上传图片能正常生成图
• 上传图片不会报错（图片被跳过）
• 切换到视觉模型（如 openai/gpt-4o 、 google/gemini-3-flash-preview ）图片能正常传入
• 风险与兼容性：仅跳过不支持的图片输入，其他 provider 行为不变；后续可抽象为统一的“视觉能力映射”

[vercel]

Someone is attempting to deploy a commit to the dayuanjiang's projects Team on Vercel.

A member of the Team first needs to authorize it.

[DayuanJiang]

谢谢贡献。
静默失败的话又会有很多人问为什么图片上传了没效果 e.g. #324
最好是把带Image_url的错误信息转换成更加可读的错误信息，例如：

next-ai-draw-io/components/chat-panel.tsx

Lines 591 to 594 in 6bb33ee

	// Translate image not supported error
	if (friendlyMessage.includes("image content block")) {
	friendlyMessage = "This model doesn't support image input."
	}

[JoeGlenn1213]

对，感觉只能维护一个 model-capabilities.json，或者 启发式判断了

[DayuanJiang]

@JoeGlenn1213 Thanks for the fix!
However, this seems like a lot of code change (215 additions) for what could be a one-line fix. How about just updating line 592 to catch the DeepSeek error message?

 if (friendlyMessage.includes("image content block") || friendlyMessage.includes('unknown variant "image_url"')) {
     friendlyMessage = "This model doesn't support image input."                 
 } 

next-ai-draw-io/components/chat-panel.tsx

Lines 591 to 594 in 6bb33ee

	// Translate image not supported error
	if (friendlyMessage.includes("image content block")) {
	friendlyMessage = "This model doesn't support image input."
	}