[Model] Add WearableMLP for short-window wearable classification

[ChanDai1214]

Contributor name: Chan Dai
NetID/email: chandai2 / chandai2@illinois.edu

Type of contribution: Model

Original paper:
https://proceedings.mlr.press/v248/kasl24a.html

High-level description:
This PR adds WearableMLP, a PyHealth-native MLP model for binary and multiclass
classification from dense wearable features. The contribution is inspired by
Kasl et al. (2024), which studies wearable-based acute illness monitoring using
shared short-window features such as resting heart rate and sleep. This model
supports short-window wearable prediction settings where multiple days of
features have already been converted into a fixed-length vector.

Implementation summary:

• Added new model file: pyhealth/models/wearable_mlp.py
• Exported the model in pyhealth/models/init.py
• Added synthetic unit tests for model initialization, forward pass, output
  shapes, and gradient computation
• Added an ablation/example script with hyperparameter variations
• Added API documentation page and updated the models index

Files to review:

• pyhealth/models/wearable_mlp.py
• tests/test_wearable_mlp.py
• examples/sample_dataset_binary_wearablemlp.py
• docs/api/models/pyhealth.models.wearable_mlp.rst
• docs/api/models.rst

Testing:

• Tests use synthetic data only
• Example script uses synthetic data only
• Model test command:
  python -m pytest tests/test_wearable_mlp.py -q

Ablation study:
The example script runs a small ablation study over hidden dimension, dropout,
and learning rate using a synthetic short-window wearable classification task.

Notes:
This contribution is a PyHealth-native model implementation inspired by the
paper’s short-window wearable prediction setting, rather than a full replication
of the paper’s complete multi-dataset benchmark pipeline.