Livermask

What's changed
- Bug fix related to model downloading from Google Drive 15
- Switched gdown for large model with custom HTTP + GitHub Releases solution.
- Improved CI to use compiled wheel instead of PyPI release during testing.

**Full Changelog**: https://github.com/andreped/livermask/compare/v1.3.1...v1.4.0

trained-models-v1
This release simple makes the pretrained models available through GitHub Releases.

This was made due to instabilities using Google Drive with large files.

Minor bug fixes mostly relevant for macOSX, but fixed a critical bug related to the vessel model.

**Changes**:
- Included yaml-file as dependency, as it was not included before
- Updated path to yaml-file to work across operating systems
- Changed multiprocessing start method to "spawn" to work with macOSX
- Livermask now works without having CuPy installed and without needing to specify "--cpu" flag

**Full Changelog**: https://github.com/andreped/livermask/compare/v1.3.0...v1.3.1

**Changes**:
- Added option for vessel segmentation (hepatic vascular system)
- Added pretrained deep vessel model
- Added Chainer and CuPy as dependencies, to support new model
- Added both CPU and GPU support for new model
- Fixed such that two models can be run sequentially, without GPU memory leakage
- Major refactoring

**Full Changelog**: https://github.com/andreped/livermask/compare/v1.2.0...v1.3.0

**Changes**:
- Bug fix related to file extension check, see [here](https://github.com/andreped/livermask/issues/10#issue-1033477092)
- Reduced memory consumption when post-processing the prediction
- Change order for correction backwards slashes to be properly used in the initial checks

**Changes**:
- Added support for batch mode (now supports providing directory as input)
- Added build as workflow, to test build on three OS for each push using GitHub Actions.
- Added alternative python wheel for install with release