[py][fix] Improved the speed of the db generation.

The problem was the search of the annotations in the eyes dataframe. Made it faster by using the filename as index and creating series instead of dataframes for each row.