Draft: converting pose and creating fake annotation for action inference

In this merge request: pose conversion and fake annotation creation are added to mmaciton dl_backend. pose conversion template: link fake annotation template: link

in livepose.bin the changes are added in case there is an action_backend in the config. With the action flag on, in the main loop, we collect frames determined by the window_length variable in the config and run video action detection on that window.