feat: Add SAM2 (Segment Anything Model) support

[ebowwa]

Feature Request

Add support for SAM2 (Segment Anything Model 2) segmentation masks.

Background

CoreMLPlayer currently supports:

• Object Detection (bounding boxes)
• Image Classification (labels)

SAM2 outputs segmentation masks (pixel-level), which would be a new visualization type for the app.

Requirements

1. Handle mask outputs - Parse SAM2's multi-array segmentation data
2. Render masks - Create SwiftUI overlay views for semi-transparent masks
3. Handle SAM2 inputs - Support point/box prompts for interactive mode
4. Mask visualization settings - Color, opacity, blend mode controls

Existing Infrastructure to Leverage

• ✅ Video/image pipeline
• ✅ CoreML loading (.mlpackage support already added)
• ✅ Coordinate transformation
• ✅ Overlay rendering system (DetectionRect)
• ✅ Model configuration UI

Technical Notes

• SAM2 outputs multi-array masks
• Need new SwiftUI view for mask overlay (similar to DetectionRect)
• Vision framework doesn't have built-in SAM2 support, so direct CoreML usage needed
• Mask data likely needs conversion from multi-array to visual format