They all do work and adafruit https://learn.adafruit.com/adafruit-voice-bonnet/audio-setup point you to the Hintak repo as the driver to use.
Multi omni directionals on all without dsp are actually pointless as you mixing out of phase by the distance between each mic and depending on orientation creating highpass filters where its often better to use a single channel.
There is an urban myth that a mic array is somehow better and in most cases without dsp the opposite is true as with alexa & google home or even the psp3 it was the high speed dsp algs that made those arrays work well.
Respeaker are atrocious with support the clones are cheap and would be only reason I would buy but a usb soundcard is cheaper and after such faff with drivers my preference.
I did get one of the above codec zero’s and the ALC & Noise gate look really good but its been sat on my desk until I can be bothered working out the myriad of alsa amixer settings.
Mems have tighter build tolerances than electrets even though electrets now are far from what the used to be, but in arrays and dsp where your calculating difference the tighter the tolerances the better.
So mems omnidirectional are prefect for dsp arrays but actually on there own a unidirectional electret has an element of directionality and noise rejection from the rear and likely a better choice if you don’t have a dsp alg to do that for you.
There is a lot of product that really poses as something more than it really is as the square 4mic has a nice pixel ring and no other further advantage and also doesn’t have audio out.
I got one and so did many but when you really analyse what you got much of it is pointless.
USB, respeaker, 4 mic and above unless you have found beamforming abd blindsource speration algs that run well on a pi take your pick, but I will argue that an electret unidirectional garnering an element of beamforming and noise reduction for one of the lowest cost is likely best currently.