Improves the way the most useful classes/methods from deeplabcut.pose_estimation_pytorch can be imported.

Additionally, makes the following improvements/fixes:

• a new dlc_torch.config.make_basic_project_config for COCO projects, to make the creation of model configuration files easier (see deeplabcut/pose_estimation_pytorch/README.md for an example)
• use get_pose_inference_runner and get_detector_inference_runner in analyze_image_folder (instead of get_inference_runners) to decouple detector and pose model loading.
• video_inference no longer needs to be given a task, as the task was not useful (if a detector runner is given, object detection will be run before pose estimation)
• load_head_weights was not correctly given to train in train_network
• Loader.ground_truth_bboxes now also returns image sizes, which makes evaluation of detectors easier