Implement StarDist deep learning segmentation plugin with optional dependencies

[Copilot]

This PR adds StarDist deep learning segmentation capabilities to ZarrNii, enabling accurate cell and nuclei instance segmentation using pre-trained or custom models.

Key Features

StarDist Plugin Implementation

• New StarDistSegmentation plugin extending the existing SegmentationPlugin base class
• Support for both 2D and 3D pre-trained models (2D_versatile_fluo, 2D_versatile_he, 3D_demo)
• Custom model loading from file paths
• Configurable GPU acceleration, probability thresholds, and NMS parameters
• Smart memory management with automatic tile size determination

Optional Dependencies Management

• StarDist dependencies are optional via zarrnii[stardist] extras
• Graceful fallback when dependencies are unavailable
• Proper import error handling with helpful installation messages

Large Image Processing

• Integration with dask_relabeling for efficient tiled processing of large images
• Configurable overlap sizes for better edge handling in tiles
• Automatic detection of when tiled processing should be used based on image size

User-Friendly API

• Convenience method segment_stardist() added to ZarrNii class
• Consistent with existing segmentation API patterns
• Support for multi-channel images (automatically uses first channel)

Usage Examples

from zarrnii import ZarrNii

# Load image
znimg = ZarrNii.from_ome_zarr("image.ome.zarr")

# Basic StarDist segmentation
segmented = znimg.segment_stardist(
    model_name="2D_versatile_fluo",
    prob_thresh=0.5,
    use_gpu=True
)

# For large images with tiled processing
segmented = znimg.segment_stardist(
    model_name="2D_versatile_fluo",
    use_dask_relabeling=True,
    overlap=64
)

# Custom model
segmented = znimg.segment_stardist(
    model_path="/path/to/custom/model"
)

Installation

# Install with StarDist support
pip install zarrnii[stardist]

# Optional: For large image tiled processing
pip install git+https://github.com/TheJacksonLaboratory/dask_relabeling.git

Testing & Quality

• Comprehensive test suite with 15 test cases covering all functionality
• Tests gracefully skip when StarDist dependencies are unavailable
• Proper mocking of external dependencies for reliable CI/CD
• All code quality checks pass (Black, isort, flake8)
• Extensive documentation with examples and usage patterns

The implementation maintains backward compatibility and follows existing ZarrNii plugin architecture patterns while adding powerful deep learning segmentation capabilities.

Fixes #56.

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.