This Pull Request makes several fixes to bugs in the ModelZoo Tab:

• adds pseudo threshold widgets back to the _build_tf_attributes
• restores showing / hiding the pytorch-specific widgets (torch_widget) on engine change
• restores showing / hiding the detector settings on engine change
• hides the detector row when humanbody supermodel is selected
• hides correctly the torch adaptation widget when superanimal_humanbody is selected

It also adds the following improvements from #3038 [outdated]:

• adds create_labeled_video argument to the video_inference_superanimal() method, and adds a "Create labeled video" checkbox
• hides adaptation settings when adaptation checkbox is unchecked
• adds batch size combo boxes for pose model and detector
• uses set_combo_items() method introduced in BUCTD GUI integration #3016 to change the contents of Qt combo boxes - this is a cleaner approach and avoids callback loops, thus improving the reactivity of the tab

It also refactors deeplabcut/gui/tabs/modelzoo.py:

• splits long methods into smaller ones
• removes code dupplication
• restores VideoSelectionWidget and adds hide_videotype option to it)