fix(report): give preset reasons a stable non-empty slug for CJK labels

ban_settings() derived each reason slug via sanitize_key(label), which
strips CJK text to an empty string; the frontend filter then discarded
every reason (including the schema defaults) and fell back to the
hard-coded list. Derive the slug from an ASCII key when possible and
from a stable md5 short hash of the label otherwise, and relax the
frontend filter to the label only. Adds a WP-stub harness that verifies
the defaults path end to end (tests/dev/verify-ban-settings-defaults.php).

tests/widget-system/report-ban-contract.test.cjs

@@ -108,6 +108,10 @@ test('ban/report settings page mirrors the reference theme schema', () => {
   // 缺键回退 schema 默认值（参考主题 _pz 语义）：运行时经 read_domain 读取，
   // 裸 get_all() 会让未保存过分区的站点拿不到 ban_preset_reason 默认值。
   assert.match(runtime, /read_domain\('users', 'report'\)/)
+  // 预置原因 slug 不得由 sanitize_key(标签) 直接派生：中文标签会被清洗成
+  // 空串，前端按 slug 非空过滤后整张原因列表会被丢弃。ASCII 键优先，
+  // 否则用标签内容的稳定 md5 短哈希。
+  assert.match(runtime, /md5\(\$text\)/)
   // 死链清理：user_report_categories 从不是真实存储键，解析器必须删除。
   assert.ok(!runtime.includes('parse_report_categories'), 'dead parse_report_categories must stay removed')
   assert.ok(!runtime.includes('user_report_categories'), 'dead user_report_categories key must stay removed')