Modularization and caching of text length widget

Removes extraneous debugging print statements

Begins modularizing so that each widget can be independently loaded without having a requirement on the ordering of load_or_preparing in app.py. This means that each function corresponding to a widget will check if the variables it depends on have been calculated yet. If not, it will call back to calculate them. Because of the messiness this causes with passing the use_cache variable around, I've now set use_cache as a global variable, set when the DatasetStatisticsCacheClass is initialized, and removed the use_cache arguments appearing in nearly every function.

Removing need to keep around base dset for the header widget; now just saving what is shown -- the first n lines of the base dataset -- as a json, and loading if it's cached.

Removing any need for a dataframe in expander_general_stats; instead making sure to cache and load the small amount of details needed for this widget. Note I also moved around a couple functions -- same content, just moved -- so that it was easier for me to navigate through the code. I also pulled out a couple of sub-functions from larger functions, again to make the code easier to work with/understand, as well as helping to further modularize so we can limit what needs to be cached.

Splitting prepare_dataset into preparing the base dataset, and the tokenized dataset. This will help us to have further control over caching and loading data, eventually removing the storage of base dataset.

Continuing cache minimizing in new repository. Please see https://github.com/huggingface/DataMeasurements for full history

:bug: filter_vocab -> filter_words

:bug: really make sure log_files/ exists

:tada: init