exclude incomplete models from comparison plot

e.g. comparing IT vs behavior, a chunk of models (efficientnet-b*) were not scored on some of the IT benchmarks, leading to a score of 0 which in turn pushes down the IT average -- I think these models with missing scores should be excluded from the comparison

related to #136

No file, test, or entry point is named. Start by locating the comparison plot and the code that turns missing benchmark scores into zeroes; inspect how IT and behavior averages are computed. Done means models missing scores for some benchmarks no longer lower the comparison averages, with the incomplete-model case covered by the relevant check.

Written by the indexing model from the issue text.