Because they are not absolute figures, just relative. They result from looking only at the axial force vector in the direction of the rider.
Total aerodynamic drag, and actual CdA of the rider-bike system increases with yaw. If we'd only rotate the rider-bike without moving the scale we'd see that.
It's just that we ignore the lateral force vector because we say the frame-wheels-tires pick it up, and it's easier to test that way. The frame-wheel-tires system is not going to be perfect at transfering this force to the ground (hysteresis losses in the system plus slippage impacting Crr). I don't think it should be totally ignored. I feel it would be better to start with total drag on the system, separate axial and lateral force vectors and work from there. It would be interesting to test Crr with a lateral force component corresponding to the load on the system from lateral force at various yaw angles and speed.

---

____________________________________
Pain is inevitable. Suffering is up to you.