Load region→lang mapping from cterdam/iso_3166-1_alpha-2

Replace hardcoded SLAVIC_REGIONS / SINOPHONE_REGIONS sets with a
session-scoped fixture that loads the authoritative mapping directly
from cterdam/iso_3166-1_alpha-2, as documented in the README.

test/test_invariants.py

@@ -24,24 +24,12 @@ import pytest
 from collections import defaultdict
 from datasets import load_dataset
 
-SLAVIC_REGIONS = {
-    'RU', 'UA', 'BY', 'KZ', 'UZ', 'TJ', 'KG', 'TM', 'GE', 'AZ', 'AM', 'MD',
-    'LT', 'LV', 'EE', 'RS', 'ME', 'BA', 'HR', 'SI', 'SK', 'CZ', 'PL', 'BG', 'MK', 'FI'
-}
 
-SINOPHONE_REGIONS = {
-    'CN', 'TW', 'HK', 'MO', 'JP', 'KR', 'KP', 'MN', 'SG', 'VN', 'LA', 'TH', 'MM', 'KH', 'MY', 'BN'
-}
-
-
-def determine_expected_lang(region_code: str) -> str:
-    """Determine expected lang from region code."""
-    if region_code in SINOPHONE_REGIONS:
-        return 'zh'
-    elif region_code in SLAVIC_REGIONS:
-        return 'ru'
-    else:
-        return 'en'
+@pytest.fixture(scope="session")
+def region_lang():
+    """Load region→lang mapping from cterdam/iso_3166-1_alpha-2 (authoritative)."""
+    ds = load_dataset('cterdam/iso_3166-1_alpha-2', split='train')
+    return {r['code']: r['lang'] for r in ds if r.get('lang')}
 
 
 @pytest.fixture(scope="session")
@@ -78,17 +66,17 @@ def test_name_matches_lang_field(dataset):
     assert mismatches == [], f"Name/lang field mismatches: {mismatches}"
 
 
-def test_lang_matches_region(dataset):
-    """lang field must match region's language rule."""
+def test_lang_matches_region(dataset, region_lang):
+    """lang field must match cterdam/iso_3166-1_alpha-2 for that region."""
     mismatches = []
 
     for i, entry in enumerate(dataset):
-        region = entry.get('region')
+        region = entry.get('region') or ''
         lang = entry.get('lang')
-        expected_lang = determine_expected_lang(region)
+        expected = region_lang.get(region, 'en')
 
-        if lang != expected_lang:
-            mismatches.append((i, entry.get('code'), region, lang, expected_lang))
+        if lang != expected:
+            mismatches.append((i, entry.get('code'), region, lang, expected))
             if len(mismatches) >= 5:
                 break