Tiramisu-brulee

-------------------

* Add AWS extras (MLFlow and `train` and `serve` console scripts)
* Add option to resample images within a subject for consistent orientation
* Add optional check of DICOM images to determine if they are uniformly sampled
* Make package compatible with Python 3.6 and 3.9
* Split CLI functions into a subpackage for better organization

-------------------

* Add MLFlow logging option
* Add support for reading DICOM images and writing DICOM (Segmentation Objects)
* Fix some type hints and make pos_weight a vector of length 1

-------------------

* Make reorientation to canonical optional
* Add option to track best network on validation Dice, PPV, loss, or ISBI15 score
* Unify and simplify the positive weight in focal/bce component of combo loss
* Change flip in spatial augmentation to only do lateral flips
* Fix predict_probability flag in CLI

-------------------

* Fix Dice score component of almost_isbi15_score metric

-------------------

* Fix reorientation to original orientation from canonical in prediction.

-------------------

* Migrate to Github actions for testing and deployment.