The diagrams reproduced in your question appear to be intended to illustrate the same, small (but non-zero) angle-of-attack in all 3 cases.
As for your quote--
Then after you increase the angle of attack and fly at that angle of
attack for some amount of time and the relative airflow changes
accordingly.
I think what you may be missing is that "flying at that angle of attack" doesn't imply flying at a constant pitch attitude. For simplicity imagine some sort of weird loop1 where we make a special point of holding the AoA constant starting from immediately after we've first "pulled G's" and began making the flight path curve. Since the direction of the velocity vector will constantly be changing throughout the loop, so will the direction of the relative wind, by the same principle that is illustrated in the diagrams in your question. So your quote above is arguably true. But AoA is constant, not decreasing.
But transitioning from horizontal flight to a constant nose-higher pitch attitude (one that eventually results in a stabilized climb) is a different situation. If we constrain the problem that way, one could argue that once the pitch attitude is established, we'll see the angle-of-attack decrease due to the changing direction of the flight path during the time interval that the flight path is still curving upwards.2 We could further constrain the problem in such a way that the angle-of-attack could indeed end up at zero, if we are flying with non-symmetrical airfoil that still generates lift at zero degrees AoA.
Exactly what we'd be doing with the elevator during this time is a different question that we can't really answer without knowing a few more constraints. (To a first approximation we'd expect to need to be moving the stick or yoke forward during the time interval that the flight path is still curving upwards, but there are many complicating effects that cause exceptions.) From a practical piloting perspective, it generally wouldn't be fair or useful to say that the mere fact that we are transitioning into a steady-state climb is the fundamental reason that the angle-of-attack is decreasing during this maneuver, or that climbing flight intrinsically tends to offer stall protection, or that climbing flight is generally conducted at a lower angle-of-attack than cruising flight. The decrease in AoA during the period that the flight path is still curving upwards can be viewed as an artifact of the fact that we've arbitrarily chosen to hold the pitch attitude exactly constant at this moment, rather than the AoA. Yet flying by reference to pitch attitude rather than purely by reference to AoA is indeed something that we actually do in many real-world instances.
For an extreme case of this scenario, forget about starting from "cruise flight" and forget about the initial increase in angle-of-attack in your quote above, and instead simply imagine that we are established in horizontal slow flight deep on the "back side" of the thrust curve or power curve, with the nose very high3, and then we add still more power (maybe we have an afterburner or two?) to accelerate without allowing the pitch attitude to change. Given sufficient power, and a non-symmetrical airfoil, we certainly could eventually end up in very high-airspeed climb with zero degrees AoA. The diagrams included in your question are clearly not intended to illustrate this sort of scenario!
And to frame the problem in yet another way, in a sustained, truly vertical climb the wing actually must indeed be at the zero-lift AoA4,5. Which would indeed protect us from stalling, as long as we constrain ourselves to manipulate the controls in such a way as to stay in that sustained vertical climb. But looking at extreme cases like that widens the discussion beyond what you seem to be asking about in your question.
So, it's complicated. We have to be very specific about which variables we are constraining when we ask questions like this. The fundamental reason an airplane stalls is that the pilot moves the control stick too far aft, thus placing the wing at the stall AoA.6
Footnotes--
It's interesting to consider exactly what such a loop would look like and how different the required control inputs would be from a "normal" well-flown loop. The answer is far from obvious, and in most aerobatic light planes we can't accurately monitor AoA directly. A good potential new ASE question?
Take care to note the importance of the word choice here-- "curving"-- a "curve" is not a linear climb at constant rate or even a linear climb with a positive rate of increase in airspeed and vertical speed. In the thought experiment here, the time that the flight path will actually be "curving" upwards as the pitch attitude remains constant (and the AoA is therefore decreasing) will tend to be brief. And don't overlook the fact that the upward curve of the flight path will typically, but not necessarily, have been preceded by an initial increase in AoA.
A delta-winged aircraft might be would be a good choice for this demonstration, due to their good flight characteristics at "high alpha".
So long as the aircraft is not tailsliding! Also, ignoring the need to offset any tail force, thrust line angle, etc.
See for example the first set of diagrams in this related ASE answer: Does lift equal weight in a climb?
This simplified argument is not intended to imply that the stick or yoke position that commands the stall angle-of-attack truly remains constant during turning flight or loops-- see for example this related ASE answer: What has happened to make me experience negative G with the control stick FULL AFT near the top of a loop?
