A note at the top - I am happy to do the PR for this, I just wanted to verify it was even something the project wanted before dropping a PR.

Is your feature request related to a problem? Please describe.
Hello! I maintain a sister project (https://github.com/sam-delap/deepxromm) for integrating DeepLabCut and XMAlab. Currently, I'm implementing my own random shuffling algorithm for frame selection, however I'd like to make better use of the existing deeplabcut.extract_frames toolkit.

However, in order to then parse the subset of data I need in XMAlab format and transform it into DeepLabCut format to create my training dataset, I either have to:

• List the contents of each video directory and parse frame numbers from the file names.
• Use the frameselectiontools util function directly and re-implement error handling and image extraction.

Describe the solution you'd like

Modify the function definition of extract_frames to add both input type hints and a return type (would still be list[int] or None given the other branches of the function) so that if frames are successfully picked using an algorithm without user input, the frames selected are returned to the caller.