Add multi-task learning for Sleep Stage Classification and ADHD Detec…

[Abhro034]

…tion

This commit adds comprehensive support for multi-task learning with GraphS4mer, enabling simultaneous sleep stage classification and ADHD detection from PSG data.

Key features:

• Multi-task GraphS4mer model with separate heads for sleep stages and ADHD
• Custom dataset loader for handling both epoch-level and patient-level labels
• Complete training pipeline with configurable loss weights
• Inference script with detailed prediction outputs
• Comprehensive documentation and example usage

New files:

• model/multitask_graphs4mer.py: Multi-task model architecture
• data/multitask_dataset.py: Data loading utilities
• train_multitask.py: Training script with early stopping and checkpointing
• inference_multitask.py: Inference script for predictions
• example_multitask_usage.py: Complete example workflow
• MULTITASK_README.md: Detailed documentation

Benefits:

• No size mismatch issues with automatic resolution calculation
• Flexible architecture supporting both S4 and GRU temporal models
• Joint optimization leverages shared representations
• Patient-level and epoch-level evaluation metrics
• Production-ready with comprehensive error handling