feat: Add combine parameter to Sam2Segmentation for flexible mask outputs

[veighnsche]

This update introduces a new combine parameter to the Sam2Segmentation node, enhancing its flexibility for segmentation tasks. The combine parameter allows users to specify whether the segmentation masks should be combined into a single tensor or output as individual masks in a batch tensor format.

Key Changes:

1. combine Parameter:

   • When set to True (default), all masks are combined into a single tensor, ensuring backward compatibility with workflows relying on combined masks.
   • When set to False, the node outputs a batch tensor containing individual masks, which can be processed independently.

2. Improved Compatibility:

   • Ensures all outputs are tensors, avoiding issues with downstream nodes like MaskListToMaskBatch expecting objects with .shape attributes.

3. Code Refactoring:

   • Enhanced segmentation logic to handle different output formats cleanly.
   • Improved handling of input masks, coordinates, and bounding boxes for consistency and clarity.

Testing and Validation:

• Tested the node with both combine=True and combine=False scenarios.
• Verified compatibility with downstream nodes to prevent errors such as AttributeError: 'list' object has no attribute 'shape'.
• Ensured output consistency and alignment with expected tensor structures.

This change adds flexibility for segmentation workflows while maintaining backward compatibility, making the node more versatile for varied use cases.

[Mithrillion]

This is actually extremely useful. With the individual masks, we will be able to automatically attach regional prompts to different areas of an image input, potentially greatly improve upscaling, style transferring etc. quality. I managed to piece together a very fragile automatic regional prompting workflow, but because extracting the text labels from the florence2 dense region caption output is still a pain, it is very convoluted. However I think if the florence2 node can also provide list-form regional caption directly, it can work wonderfully with the functionality added here.