Somehow, I don't think any new AI features Apple does end up announcing at WWDC will be limited solely to the M4 chip. Apple is keenly aware that upgrade cycles have slowed, nobody is going to run out tossing a perfectly good 1-2 year old device just for said functionality, and my guess is that these features will extend all the way back to the M1 chip.
After all, you want higher adoption so as to incentivise developers to incorporate those features into their apps. An M1 iPad that was purchased back in 2021 is good till 2026.
The reason why the airs didn't get the M4 chip is (my best guess) because they don't need them. My understanding is that the new iPad pros needed the custom M4 chip to power that tandem OLED screen, and maybe for the better thermal efficiencies since they are going into thinner devices. The iPad Air essentially recycles the previous iPad form factor, so M1 or M2 suffices.
I don't know if it's a cost saving measure as well. Would the M4 chip cost more to manufacture than the M2 chip at this juncture?