Support multiple dataframe in preprocessing

Here are the changes implemented by this MR

• The preprocessing was cleaned up (I forgot to remove a few things in the previous branch)
• A new subdirs parameter in the preprocessing configuration, that can be used to have several dataframes as input
• A preprocessing step that loops over the input dataframes.
• A preprocessing step that uses groupby to loop over events, which is faster than filtering the events at every iteration
• A new "sorted-wise edges" that links the hits by sorting the hits by increasing z
• An updated "plane-wise edges" that links the hits that belong to adjacent planes within a particle
• A new training that is being slowly set up (track-weighting)

I tried to implement some multi-processing in the preprocessing and but seemed to be actually slower so I gave up on the idea.