Add LaMa Core ML model for AI-powered inpainting

- Add LaMaFP16_512.mlpackage (~90MB) for high-quality object removal
- Add LaMaInpainter.swift wrapper with image preprocessing and merging
- Modify InpaintEngine to use LaMa first, gradient fill as fallback
- Fix brush mask size (use scale 1.0 instead of screen scale)
- Fix LaMa output size (use scale 1.0 in merge function)
- Add model loading wait with 5 second timeout

The LaMa model provides significantly better inpainting quality compared
to the gradient fill method, especially for complex backgrounds.

.gitignore

@@ -54,3 +54,4 @@ logs/
 # Temporary files
 *.tmp
 *.temp
+inpaint-ios-reference/