Training dataset generation and training code upgrades

To make DAN training easier, we first need to improve the codebase.

• Import and format data from Arkindex
• Update training code

changed milestone to %ML Prod - October 2022 n°2

added P2 label

assigned to @starride

mentioned in issue #2 (closed)

marked this issue as related to #2 (closed)

mentioned in merge request !4 (merged)

mentioned in merge request !5 (merged)

closed