- Pass auth token to schema discovery and descriptor generation in
  tg-load-structured-data, fixing 401 errors with IAM enabled
- Fix row query pagination: replace single-page async_execute with
  async_scan that streams pages and applies filters without
  materialising the full result set (OOM on large datasets)
- Add missing filter operators (not, startsWith, endsWith, not_in)
  to row query post-filter matching
- Fall back to scan path when an indexed field is queried with an
  empty string value, since empty index values are not stored
- Revert top-level indexes array support — the current table schema
  overwrites rows with duplicate index values, so only primary_key
  fields are safe to index until the schema is redesigned