Add pre-processing scripts as per the Datasheet's instructions

[AndreFCruz]

We should add python scripts for the data's pre-processing according to the datasheet's instructions.
For example:

• split train/test according to the month of the year;
• binarize sensitive attribute column (according to the datasheet and paper this is done with <50 and >=50 years old);
• correctly load data from the parquet files (need to list pyarrow in the requirements.txt file)
• helper functions to label encode (or one-hot encode) categortical columns;
  • their current string encoding is incompatible with some well known algorithms;
  • or use dtype="category" if using pandas, this solves the problem for some algorithms including LightGBM;

Ideally this would all be implemented in a baf_helper python package, which could deal with downloading the data as well.

NOTE:

• we can use this package to download files from GDrive
• we should also summarize the readme and be more straight to the point on how to get usable data to train ML models;