use more generic binary series loader #362

Currently to read distributed series binary data we use the spark sc.binaryRecords method. This depends on Hadoop, and may have unexpected ordering behavior across different systems. We should probably just write a method that maps the list of file names and reads the files directly, just as we read binary images.

cc @jwittenbach