[CS598 DLH] Add DREAMT OSA classification

[ssrir5]

Title

Add DREAMT dataset + OSA classification task

Summary

This PR adds support for the DREAMT sleep dataset and a simple patient-level OSA classification task based on DREAMT clinical metadata.

Changes

• DREAMTDataset

  • Wraps PhysioNet DREAMT (1.0.0–2.1.0) as a BaseDataset.
  • Builds dreamt-metadata.csv from participant_info.csv.
  • Normalizes columns (e.g., SID → patient_id, AHI → ahi, etc.) and attaches paths to 64Hz/100Hz CSVs.

• DREAMTOSAClassification

  • New BaseTask for patient-level OSA prediction.
  • Tasks:
    • ahi_severity_4class (normal / mild / moderate / severe).
    • ahi_binary_15 (AHI < 15 vs ≥ 15).
    • oahi_binary_5 (OAHI < 5 vs ≥ 5).
  • Uses tabular clinical features (age, gender, BMI, SaO₂, arousal index, history, disorders).

• Tests

  • tests/core/test_dreamt_osa_task.py:
    • Checks label mapping and feature dict for ahi_severity_4class.
    • Verifies split handling and binary label for ahi_binary_15.