Read data from vector tile with low memory approach

[springmeyer]

Flagging that this will be viable to consider once these upstream tasks are done:

• optimized vector_tile_datasource that works directly on protobuf buffer mapbox/mapnik-vector-tile#92
• Stop storing vector_tile::Tile inside mapnik.VectorTile object #419

[springmeyer]

What is done:

• mapnik-vt now has a solid and well tested datasource implementation that works - in a low memory way - against the raw protobuf buffer and does not need an intermediate libprotobuf object (optimized vector_tile_datasource that works directly on protobuf buffer mapbox/mapnik-vector-tile#92)
• pbf.hpp has been re-named to protozero (https://github.com/mapbox/protozero) and has 100% test coverage and a new writing interface
• node-mapnik branch (https://github.com/mapnik/node-mapnik/compare/parse-no-more) is in motion to start leveraging the performance gains from all this.

Next steps:

• port vtile.composite to use tile_datasource_pbf
• port vtile.render to use tile_datasource_pbf
• port vtile.query to use tile_datasource_pbf
• port vtile.queryMany to use tile_datasource_pbf
• upgrade node-mapnik and mapnik-vt to the latest protozero release when it lands: First tag mapbox/protozero#4
• port vtile.isSolid impl that works on raw bytes - is_solid impl that works on raw bytes mapbox/mapnik-vector-tile#140

Would also be nice to hit eventually (but not performance critical):

• port vtile.toGeoJSON to use tile_datasource_pbf
• port vtile.toJSON to use tile_datasource_pbf

Longer term, be able to completely remove libprotobuf dep:

• add validation function that uses protozero
• port vtile.addGeoJSON to use protozero writing
• port map.render(vtile) to use protozero writing
• port vtile.toJSON to use protozero reading

[springmeyer]

status update:

• all tests passing
• protozero now used in all performance critical code paths
• therefore basically ready to:
  • test with more deps (will require usage updates to avoid vt.parse())
  • consider merging based on the outcome of testing with tilelive-bridge/tilelive-vector, etc
• for the full perf benefit will be great to hit Implement gzip compression in the C++ layer #456 next and get it updated to protozero code.

[springmeyer]

closing this, will now track at pull that implements the majority of what we need for performance critical paths: #505