Be more careful to retry when the feature count is exceeded

[e-n-f]

This fixes a regression introduced in #227, where, because the ratio of base features to multiplier features can shift slightly over the course of a tile, a tile might exceed the feature count limit during the assembly of features but then might be judged to be below the limit by the end of the tile, causing the tile to be truncated rather than retried.

Now it explicitly checks when any features have been excluded for being over the limit and always retries the zoom level if any dynamic dropping options have been specified.

There was also a bug where features that were dropped onto a multiplier cluster rather than dropped entirely from the tile were not counted as having been dropped and did not prevent the zoom pyramid from being truncated. Such features are now counted as dropped_as_needed and prevent zoom pyramid truncation.

In addition, it changes the maxzoom reported for tilesets generated with --generate-variable-depth-tile-pyramid to be the highest zoom level at which tiles were actually generated, not the originally-specified maxzoom.

It also changes the order of operations in tippecanoe-overzoom to clip features to just those that will be included in the output tile before doing any work related to filters or multiplier clusters.

Finally, it allows the value of a BUILD_INFO environmental variable to be included in the --version output and generator metadata.

[migurski]

Approved based on our 1:1 conversation