A feedback cancelling solution would probably increase latency however, so it would kill the 'low latency' claim if it was on by default.
Definitely something needed to be switched on for problematic setups though, since the disadvantages would be outweighed greatly (not having feedback \ echo is a necessary feature).
It shouldn’t necessarily, you can find out how much of the speaker output ends up in the microphone input and compensate. I’m sure the models are more complicated to deal with echoes and distortion but the same approach should basically work: fit the filter offline, apply it online.
