How is this different from a random passerby identifying the suspect from the grocery store footage, or a police officer already knowing the suspect? They are just clues until any proof is found (ex. a gun that matches the bullets).
A judge probably would not rule in favor of a warrant based solely on a tool whose utility is entirely unproven and which itself warns that it should not be relied upon as evidence, while they might rule in favor of eyewitness testimony.
The main problem in this case was the fact that the police misrepresented their evidence to the judge. It is theoretically possible that a judge may have issued a warrant on the basis of the facial recognition tool, in which case the evidence from the search would not necessarily need to be thrown out. If that were the case, I'd expect the defense to appeal the validity of the warrant, in which case all the questions you're asking would come into play.
If Clearview AI is indeed functionally equivalent to some random passerby, then what value-add does it actually have over that baseline? Either it's really the same and there's no value to justify deploying it. Or there's something else at play (say, scale) that's worth examining on its merits and risks.
So the warrant was granted from false pretense.