Phenopype

-----
bugfix:
- adding image files in "link" mode to a projet will now use relative links instead of absolute
new functionality:
- project.create_training_data(): exports a set of annotations that can be used for training machine learning models. The method extracts a specific set of annotations from the entire project (for a specified tag) and formats them in a specific way so that the data can be immediately processed by a given architecture or framework. The first implementation is for "ml-morph" (https://github.com/agporto/ml-morph / Machine-learning tools for landmark-based morphometrics), a Python toolchain for automatic landmark detection in images.

-----
bugfix:
- detect_mask (formerly detect_shape) would return faulty masks
- remove debug markers ("test", "bier", etc.)
new functionality:
- project.export_zip(): compress the current project and store as a zip archive to the root folder (option to exclude imgs)
- detect_shape becomes detect_mask
- more legacy conversion added (function names in Pype config files will be automatically renamed to new syntax)

-----
- removed numpy/pandas version selection in setup.py to reduce conflicts with other packages and environments
- zooming during Pype has now memory: it will keep the last zoomed in area focused (except the final inspection window)

-----
- edit_contour is now directly applied to binary image when used in Pype (accumulative)
- improved pype name check

-----
new major release - breaks all existing versions

too extensive to document completely, but in a nutshell:
- we now have annotation functions which save image annotations (contours, drawings, data) to a dict (internally) and are saved as json
- GREATLY reduced boilerplate code allowing more flexibility when introducing new functions (for developers and contributors)
- better code annotation and more settings
- templates externalized
- all new docs!!

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

-----
bugfix:
- save_masks as df would take ndarray instead of DataFrame
- select_canvas with ndarray would not allow to select raw
- draw_landmarks now takes label=False/True and offset argument