0.25.5

New Features 🎉

• Predicates that cannot be answered by the ParadeDB index can now be evaluated by intersecting with a bitmap scan over another PostgreSQL index (btree, GIN, or GiST), rather than checking every matching row against the heap, reducing buffer reads by up to 20x for selective filters.
• Replaced the paradedb.check_aggregate_scan and paradedb.check_topk_scan GUCs with a single unified paradedb.planner_warnings GUC (off, warning, or error). When set to error, queries that cannot use an optimized ParadeDB scan raise an error instead of logging a warning.

Performance Improvements 🚀

• MPP search readers are now initialized only in the processes that execute them, avoiding redundant index opens on the leader and opening fast fields lazily per segment.

The full changelog is available on the GitHub Release.