Change peptide filtering to never violate combined_score order

[iskandr]

Currently we generate many sliding windows around a variant and then filter them to only keep those that are within 1% of the max combined_score. In cases where two variants have very similar combined scores we may actually use a peptide for the higher ranked variant with a lower epitope score than a peptide from the next variant. To fix this we should change core_logic.py to first find the max combined score for each variant and then have a variable amount of tolerance in peptide filtering based on the difference between the current variant's score and the next variant's score.