Bump Tantivy to 0.22 and also the root dependencies we share with it.

This does not yet make use of the new options around Tantivy's Document trait but just continues using the existing options (also under their old names instead of the new ones).

As a follow-up we can optimize indexing by implementing Document and Value directly on top of Dataset instead of building up owned field-value pairs.

Note that deploying this will also require us to rebuild our GeoNames index which also uses Tantivy under the hood.