False positives in tilestats

Especially with https://github.com/mapbox/tippecanoe/pull/553 moving evaluation of conditionals from parsing time to tiling time, layers and attributes can make it into the tilestats that never actually appear in the tiles.

I'm not sure what can be done other than to move tilestats generation to generally later in the process, which is slower and more prone to reporting the same feature multiple times (when split across tiles and duplicated across zoom levels) than doing it during parsing is.

Start by tracing tilestats generation and the move of conditional evaluation from parsing time to tiling time. Determine how to exclude layers and attributes that never appear in output tiles while avoiding duplicate reporting from split features and repeated zoom levels; done when false positives are eliminated without unacceptable duplication or slowdown.

Written by the indexing model from the issue text.