enhance: add original image handling and return types in BizyAirSegmentAnythingPointBox

[lcolok]

Workflow:
Fill+Redux+SAM（原作者：Lingyuzhou）.json

This PR improves the BizyAirSegmentAnythingPointBox node by adding an additional output that passes through the original input image, optimizing workflows that involve FLUX Fill + Redux.

Previous Behavior

• The node only output the processed image (with alpha channel) and the mask
• Workflows like FLUX Fill + Redux required uploading the same image twice:
  1. Once for the SAM processing
  2. Again for the actual fill operation

New Behavior

• Added a third output that passes through the original input image
• The original image is efficiently cloned and passed through without unnecessary conversions
• FLUX Fill + Redux workflows now only need to upload the image once, as the original image is available directly from the SAM node

Benefits

• Reduces redundant image uploads in complex workflows
• Improves workflow efficiency and reduces resource usage
• Maintains image quality by avoiding unnecessary conversions

Implementation Details

• Added "IMAGE" as third return type with "original_image" return name
• Uses efficient image cloning to pass through the original image
• No breaking changes to existing functionality

---

工作流:
Fill+Redux+SAM（原作者：Lingyuzhou）.json

SAM.Pick.mp4

本次PR优化了BizyAirSegmentAnythingPointBox节点，增加了原图透传输出，以更好地支持FLUX Fill + Redux等工作流场景。

现有问题

• 节点目前只输出处理后的图像（带alpha通道）和mask
• 在FLUX Fill + Redux等工作流中需要重复上传同一张图片：
  1. 首次上传用于SAM处理
  2. 再次上传用于实际的填充操作

改进方案

• 新增第三个输出用于透传原始输入图像
• 通过高效的克隆方式实现原图透传，避免不必要的格式转换
• FLUX Fill + Redux工作流现在只需要上传一次图片，可直接从SAM节点获取原图

优化效果

• 减少复杂工作流中的重复图片上传
• 提升工作流效率，降低资源占用
• 避免不必要的图像转换，保持图像质量

实现细节

• 在返回类型中新增"IMAGE"作为第三个输出，返回名称为"original_image"
• 使用图像克隆实现高效的原图透传
• 保持向后兼容，不影响现有功能