make word count computation part of the import

Accurate word counts are now essential when using
the ICU tokenizer and don't hurt for the legacy one.

Adds about an hour import time.

nominatim/clicmd/setup.py

@@ -125,6 +125,8 @@ class SetupAll:
                 freeze.drop_update_tables(conn)
         tokenizer.finalize_import(args.config)
 
+        LOG.warning('Recompute word counts')
+        tokenizer.update_statistics()
 
         webdir = args.project_dir / 'website'
         LOG.warning('Setup website at %s', webdir)