[WIP] Wesad contrastive eda

[msaunders804]

Summary

Reproduces "Contrastive Learning of Electrodermal Activity Representations
for Stress Detection" (Matton et al., CHIL 2023) as a full PyHealth pipeline.

Changes

• pyhealth/datasets/wesad.py: WESAD dataset class with EDA windowing and
  subject-level loading
• pyhealth/tasks/stress_detection.py: Stress detection task with LNSO
  cross-validation split support
• pyhealth/models/contrastive_eda.py: Contrastive EDA encoder with SimCLR-style
  pre-training, NT-Xent loss, and EDA-specific augmentations
• examples/wesad_stress_detection_contrastive_eda.py: Full pipeline example
  with augmentation ablation study

Ablation Results (1% labeled data, 5-fold LNSO)

Augmentation Group	Mean Balanced Acc	Std
Full	[from Colab run]
Generic only	[from Colab run]
EDA-specific only	[from Colab run]

Tests

• 7 dataset tests
• 7 task tests
• 21 model tests
  All passing.

Paper

Matton et al. (2023). Contrastive Learning of Electrodermal Activity
Representations for Stress Detection. CHIL 2023. PMLR 209:410-426.