Improvements in the datashuffler

For the time being the data shuffler fetch the whole dataset in memory. This is not the best scenario. Study how to do PREFETCHING (https://www.tensorflow.org/versions/r0.10/how_tos/reading_data/index.html#creating-threads-to-prefetch-using-queuerunner-objects)
Add the possibility to shuffle text lists and load them in the get_batch function.