However, I see 426mV on the optocoupler when it should be 0V.
I assume you mean 426mV across pins 1&2 of the optocoupler. If so, this shouldn't be a problem since 426mV is not enough to light the LED in the opto and turn its TRIAC on.
On high, I see 1.15V which is the forward voltage of the LED, so all good. I'm not sure if this will cause any issues with the optocoupler.
I removed R19, and the optocoupler input is 0mV. What's the trick here? If there is a simple mod for the existing PCB it would be good.
The reason for the 426mV across pins 1&2 of the optocoupler is because R19/Q1 is loading the output of IC2B (pin4) when it goes high.
This means pin-4 only goes up to 4.574V (5V - 426mV assuming a 5v supply)
Like I said, it's not really a problem, but if you changed Q1 to a MOSFET and eliminated R20 this "problem" would go away.
Another option would be to increase R19 to 1.2k (assuming a worst case beta for Q1 of 20) which would reduce the current draw from IC2B. (pin4). This would reduce the voltage to something lower than 426mV.
Also does the relay transistor need a pull-down?
If the output of IC2B (pin4) doesn't go lower than apx 700mV when it's low, then Q1 will not turn off. R19 & R20 can form a divider so that the voltage on Q1's base is less than 350mV when pin-4 goes low.