Build a database of done experiments to not relaunch existing inferences

Take into account :

• model
• database
• transformation
• version of code (to which granularity ?)

Ensure we have proper primary key (absolute img_path?) in main table that is present as foreign key in other tables to join them successfully.