This Pull Request adds Pytorch support for the create_pretrained_project() method (and closes #2732)

Specifically:

• create_pretrained_project() becomes a compat.py-like method, and calls the specific method implementation based on the specified engine parameter.
  The old create_pretrained_project() is renamed to create_pretrained_project_tensorflow() and a new create_pretrained_project_pytorch() method with the Pytorch-shuffle logic is added.
  The existing usages of create_pretrained_project() are also adapted.
• Both create_pretrained_project_* implementations call _create_training_datasets_metadata().
  This method has been cherry-picked from [WIP] DLC 3.0 - Review documentation #2881(where a first version, _create_and_save_fake_metadata(), has been implemented as a temporary fix), and adapted to work with both Tensorflow and Pytorch. It creates the metadata.yaml file in the training-datasets directory, and fills it with a first shuffle corresponding to the downloaded modelzoo model.
• The video processing logic (that is optionally done after the pretrained project has been created) is extracted to
  _process_videos(), and called by both create_pretrained_project_* implementations.
• A new individuals parameter is added to create_new_project() and to create_pretrained_project(), so that the list of individuals can be automatically filled in the different configuration files of the new project.
• An integration test script for create_pretrained_project(engine=Engine.PYTORCH) is added.

This Pull Request also does some fixes to issues that have been spotted during its development:

• Fixes list_videos_in_folder() : in original (Tensorflow-implementation) methods (f.ex. analyze_videos()), videotype argument's value by default was "" (instead of None), and this default argument value was not treated properly.
• get_super_animal_project_cfg() was using read_config() to read the superanimal project configuration template embedded into the DeepLabCut repository, which then modified the file in the DeepLabCut codebase. Now, the method reads the template with read_config_as_dict(), which doesn't modify the file.
• The previous fix created a circular-import issue, that has been resolved by moving the read_config_as_dict(), write_config() and pretty_print() methods to deeplabcut.core.config_utils.py module.
• The other superanimal integration test scripts (testscript_superanimal_adaptation.py, testscript_superanimal_inference.py and testscript_superanimal_transfer_learning.py) have been fixed.