[fix] sparsity of country tags should be enough for language address training data

scripts/geodata/osm/osm_address_training_data.py

@@ -707,9 +707,6 @@ def build_address_format_training_data_limited(language_rtree, infile, out_dir):
                 have_country = True
             _ = value.pop(k, None)
 
-        if have_country and random.random() < 0.8:
-            have_country = False
-
         if not value:
             continue