Fair Infinitesimal Jackknife: Mitigatingthe Influence of Biased Training Data Points Without Refitting