Heartkit

Release Highlights:

* Converted datasets now being pulled from S3
* Removed task specific logic from dataset classes.
* Update task structure to separate methods (e.g. train.py)
* Add helper functions to create dataset pipelines.
* Add initial foundational model to perform contrastive learning across ECG leads.

Release Highlights:

* Add augmentation pipelines.
* Add Physiokit routines for HRV metrics and R peak detection.
* Reduce model sizes by over 50% while preserving accuracy.
* Add new QTDB dataset.
* Reduced sampling frequency from 250Hz to 200Hz.
* Update preprocessing to apply biquad filter forward backward to maintain zero phase.
* Add quantization-aware training (fine-tuning) for segmentation model.
* Add MobileOne (1D variant) model architecture.
* Add additional structs to reduce shared globals in EVB code.
* Update to TF 2.13.
* Update to Python 3.11.
* Various bug fixes and performance improvements.