This PR makes three improvements to feature dropping behavior:
When dropping features as needed because a tile is too big, the decision about which features to keep or drop is made for an entire multiplier cluster, not for individual features. The cluster is either kept or dropped as a unit.
--drop-fraction-as-needed now makes decisions that are consistent between tiles and zoom levels and that follow the same progression as zoom-level-based point dropping
The features within each multiplier cluster are also sorted according to this same standard, for greater consistency between filtered features across zoom levels.
(This last behavior is still not good enough for complete consistency near tile boundaries, since sometimes features that are part of one cluster at a low zoom level will be split between tiles at a higher zoom level, which will have to be addressed in a followup PR)
This PR makes three improvements to feature dropping behavior:
--drop-fraction-as-needed
now makes decisions that are consistent between tiles and zoom levels and that follow the same progression as zoom-level-based point dropping(This last behavior is still not good enough for complete consistency near tile boundaries, since sometimes features that are part of one cluster at a low zoom level will be split between tiles at a higher zoom level, which will have to be addressed in a followup PR)