[TESTING] 4x faster polygon decoding

[springmeyer]

This speeds up polygon decoding + clipping + encoding by 4x. Polygons are going to be the slowest to overzoom, so this is significant.

The speed boost comes from avoiding the allocation of intermediate vectors to store rings. Instead we collect bbox's while we decode, only clip geometries that are not fully within the target tile, and when we have results we encode them directly to the new feature.

Current master, for test 11 gives:

node bench/bench.js --iterations 500 --concurrency 1 --package vtcomposite
11: tiles completely made of polygons, overzooming and lots of properties ... 166 runs/s (3015ms)

This PR gives, for test 11, gives:

node bench/bench.js --iterations 500 --concurrency 1 --package vtcomposite
11: tiles completely made of polygons, overzooming and lots of properties ... 728 runs/s (687ms)

Next steps:

• Review from @ericfischer
• Review from @flippmoke
• Dev package + test in staging
• determine it is ready to merge + officially release

/cc @mapbox/maps-api @artemp @flippmoke @ericfischer @millzpaugh

[artemp]

The bench results here are bogus and 4x speed-ups are due to the bug #95.

With #95 fixed I'm getting roughly equivalent results:

Current master

11: tiles completely made of polygons, overzooming and lots of properties ... 252 runs/s (1984ms)

This PR

11: tiles completely made of polygons, overzooming and lots of properties ... 251 runs/s (1992ms)

/cc @mapsam @flippmoke @ian29