GeoParser suggests a greedy parsing where geo data items are created
directly once according tags are encountered. This does not work well
for OSM where e.g. multiple tags must be evaluated at the same time or
a relation changes the semantics of a way. A couple of hacks and some
strange behavior resulted from that, e.g. converting between line
string, linear ring and polygon at some later stage or creating
multiple points of interest on top of buildings.

The simplified parser does one pass now where it extracts the
information and then interprets them. It is currently deactivated
by the USE_OLD_OSM_PARSER switch in OsmRunner.cpp in order to easily
compare both parsers while finishing it. Things left to do:
- check that osm placemark data is registered consistently
- check that multipolygon registration is possible at all
- extract non-trivial multipolygons (old parser is buggy here, new one
  tries to ignore non-trivial ones for now)
- decide how to render additional POI categories for buildings. Old
  parser created multiple POIs on top of buildings, but including POI
  icons and maybe labels directly inside polygon rendering might be
  better.