Functioning UKB RAP data prep script

[LArnoldt]

The provided script for data preparation on the UKB has some major issues:

(1) The performance issue is caused specifically by the for i in range(len(fo_fields)): loop that repeatedly executes df.select(...).where(...).withColumn(...).union(...). This performs ~1000 separate Spark jobs with repeated scans, UDF execution, and growing union lineage, which prevents effective query optimization and makes the runtime explode with column count.
(2) The vocabulary number deviates from the original implementation.
(3) The events are not correctly tokenized.

I am providing a script, which solves these issues, while basically replicating the original data preparation script example_ukb_to_bin.ipynb.