How an aircraft aerodynamically moves though the airmass is essentially an engineering problem which is very well understood. Computers solve these engineering problems very well, better than humans, and it is clear that automation of the nature used by Airbus has saved many lives. Computers, as well as pilots, require sensors to fly and if the aircraft sensors are working it makes sense for the aircraft to fly the aircraft instead of the pilots. When all aircraft systems are working, it also makes sense for the aircraft to override all 'flying' actions that the pilots make which the equations for flight deem to be dangerous. i.e. Normal Law.
When the aircraft sensors fail, as they appear to have done here, it is essential that the aircraft hands over control of the flying of the plane to the pilots. This appears to have been done properly. As the aircraft had degraded sensor information it removed any protections that rely on this sensor information. This is also a correct action to take, aircraft in ALT 2, no AoA protections in place.
It then appears that the pilots mismanaged the further flying of the aircraft. Unusual, but the evidence we have so far points strongly in this direction. The reported sidestick position invalidates all the other improbable theories proposed so far.
On the descent, it appears that the input from the aircraft sensors became valid again. This is entirely to be expected if the initial problem was icing. The aircraft would presumably have known that all it's sensors were now working. However it was latched into ALT 2, AoA protections are not in place, and so it could do nothing about the stall.
If once the aircraft had validated the sensor data, it could have legitimately reverted to normal law, it could then have applied AoA protections and all on board saved.
So why is this not programmed to happen?