I'm using a couple of TI CD4082BE AND gates to bring the interrupt pin low when any button is pressed, aka one of the rows goes low. The AND gate works.
That is not correct!. You will never get that result from an AND. Now, if you had stated I use a CD4082BE negative OR gate you would have been correct.
Calling a negative OR an AND will lead to confusion "some day"

Regarding the input pin:
I thought that being Push-Pull or Open-Drain didn't matter when a pin is used as an input. That doesn't APPEAR to be the case here. The pin I'm using for INT0 is configured as Push-Pull in the crossbar. This works great If I go around the gate
If an input pin is configured as push-pull one on these things will happen
a) the signal driving the input can not overcome the push-pull = nothing or a blown/shotlived transistor in the driving chip.
b) the signal driving the input can overcome the push-pull = it "works" while shortening the life of a/the chip(s).
c) the signal driving the input overcomes the push-pull = either the uC or the driving chip blow a transistor.
NOTE: this is true for all SILabs derivatives (e.g. f12x), I have long ago abandonede the stupid SILabs Deviates (f3x, f4x).
I guess that the direct connection has result b) and the negative OR has result a)

Now for a thing re keypads where I learned a lesson: It is "natural" to configure the outputs to a keypad as push-pull (they ARE outputs) and it works with one caveat if two keys are simulataneously ypu may connect two push pull outputs at different states togeteher with the above results a) b) or c)

Erik