Remove multiprocessing to avoid concurrency.

While cross-checking the RED data we have noticed that some runs contain more frames than expected. From the metadata in one of such red data files it was evident that reduction instructions were missing for certain number of trains. Most probable explanation of this issue is possible concurrency in multiprocessing with ThreadPoolExecutor.

Replacing multiprocessing with a simple loop results in about 1 to 3 seconds of extra computation time for a run, which seems to be acceptable.