Harnessing machine learning for accurate treatment of overlapping opacity species in general circulation models