"ml-probability-high" labeled issue turns out to be a valid one

[softvision-sergiulogigan]

We ran into a contradictory labeling situation on a valid "login" related bug where the issue was filed by a user (webcompat/web-bugs#46318) which was not labeled with ml-probability-high and another issue filed anonymously (webcompat/web-bugs#45304) which was labeled (ml-needsdiagnosis-false and ml-probability-high)

[johngian]

I think this is part of the feedback threshold that we put. I don't think we can ensure that all our predictions are 100% correct, but we can improve things by retraining the model periodically.