Also consider tokenized matches in GeoNames if they are unique.

Some place names cannot be matched using string comparisons but still yield unique matches when performing a search based on the individual token. This is added as a fallback strategy here.

This does imply a significant size increase for our index over the GeoNames databaes from 0.8 to 1.4 GB which does seem bearable as it is a one time cost though. (And it includes making name and alt. names fast fields which should speed up resolving identifiers into place names measurably.)

Closes #168 (closed)