That's the most simple way to handle it:

If we would have four years long network difficulty regression - then it's emergency, and new code handling such danger - should delay halving to the next halving, until difficulty will recover

if there is no such emergency situation - there is no trigger, and old and new code would work together like a charm = so there is no hard fork at all

simple, conservative (and beautiful) solution in my opinion, so it fits to Bitcoin very well

such solution above is fueled 100% by free market only and the natural inflation level is set by Bitcoin itself, on some certain, completely unpredictable level (and that's why I'm not sure we would be able to apply any math/model here)