I think there may be two "latch-up" modes being discussed
here. One is 4-layer (SCR) latchup between the power rails.
An input exceeding the VCM limit doesn't cause this but one
exceeding the rails by a Vbe (and with enough trigger
current behind it) might. This should be spec'd in the Abs Max
tables and defeated by proper I/O pad and attached-device
design (guardrings and such).
The other looks to be a functional / behavioral deal, a
latched circuit state but without excess / destructive
supply-supply current, stemming from the front end
pair being choked off. Both PNPs need to go above
(Vdd-Vbe-Vce(sat)) to choke the front end currents
off. One is pinned to the 2.6V reference on the right
hand amp, so this would not occur (but is pretty
close, maybe too close for low temp comfort). As
long as one input remains below that threshold the
amplifier will have the right sense because currents
will steer appropriately.
Any op amp or comparator with a phase inversion
anywhere in the Vcm range should be shunned, of
course. And I prefer that (no phase inversion) to
encompass the supply rails range regardless of
whether common mode range for performance,
does or not. But it can be hard to pick out from a
datasheet, and if the part acts wrong in this respect
it might just not be mentioned.