Refactor face detection

[framp]

Can we abstract face detection / face extraction out of these specific workers?

This would allow us to:

• deduplicate the code
• add different models other than insightface (non commercial licence)