use two diodes to make a diode "OR" gate. Connect one to the switch and one to the output of the ESP.
That was my initial thought, but Ian.M said, "You'll have difficulty meeting the L9110S input threshold requirements if you use diodes. Noise immunity will certainly suffer."
However, the input threshold is 2.5V and if I use a Schottky
such as this which has a Vf Max of 340mV @ 1A (and even more at these lower currents) which gives me an input minimum of 2.84V. The output from the ESP is 3.3V, which gives 0.46V of buffer, at least.
That leaves me with the second comment of noise immunity. I don't know if I'm vulnerable to this? The motor isn't driven by PWM; it's just full on, full off. Does running at the low end of the L9110S input voltage range mean that my motor will run at a lower speed? Both the GPIO and the motor are rated for 3.3V.
A concern with an OR gate the manual override can set one side of the bridge to high and the other side can be set by the ESP as well, leading to both being high at the same time. This is
generally not recommended with H-bridges but the truth table
in the datasheet doesn't seem to have a problem with it. It describe a situation with both high and the output is both are low.
I should have posted a schematic. Will update the OP.