Micro-optimizations of search score handling.

[isoos]

• Part of the split of Reduce the memory allocation during search with mutable IndexedScore. #8225.
• PackageNameIndex.search (used only in tests) is updated to update its filterOnPackages set on subsequent iteration. If this would be used elsewhere, it would probably improve the search by reducing the number of ngram-similarity calculation done by the index.
• PackageNameIndex.searchWord is updated to not insert entries below the 0.5 score threshold (so we don't need to update the map to remove them later).
• PackageNameIndex.searchWord also removed scores that were half of the largest score present, but in practice, with the combination of the 0.5 threshold (above) this does not have any further effect, as the maximum it would have removed would be 0.5 anyway. Removing that part is a no-op.
• The iteration change in the TokenIndex constructor spares us the internal lookup.