Turn a 2D mask drawn on one sensor frame into a labelled 3D selection and a survey part (Segmentering view, spec B2).

Lives in core (not vision) for the same reason frame_visibility does: the GUI server (ruxd_api_lib) links reusex_core but not reusex_vision, and the existing batch projector vision::project() drags in RTABMap. The projection here is a plain pinhole model plus an occlusion test against the frame's own depth image — no z-buffer, no RTABMap, no PCL in this header.