update to concat frames

[clewis7]

for now the solution for feature extractor inference loading is to just not use multiprocessing for loading in frames until I can investigate it more and figure out how to make it work with two videos where you are not necessarily getting the same frames from the workers

but, I tried out inference with sample data and it worked wonders...going to try with yuechen's data as well

[clewis7]

need to think more about allowing multiprocessing frame reading for feature extractor inference....I think I can figure it out just need to understand more about how pytorch IterableDataset works