UPDATE: I think I understand the problem. At the core it seems that face tracking doesn't exists through the rear camera system. According to one source the rear view camera physically differs from the front camera.[1] Thus it makes sense why entities with sceneUnderstanding of entityType .meshChunk gets added by the session but not .face when using ARWorldTracking. A natural solution(one also suggested by the source mentioned before) may be just to create a 3D object that gets added to the configuration using detectionObjects.[2]
Good luck!
[1] https://discussions.apple.com/thread/254147306?sortBy=best
[2] https://developer.apple.com/documentation/arkit/arworldtrackingconfiguration/2968177-detectionobjects