o Bloom filters on equality-key columns: LARGELY A NON-ISSUE (verified).
  DuckDB's COPY TO writes bloom filters BY DEFAULT and selectively per DuckDB's
  own cardinality heuristic (e.g. a low-cardinality integer key gets one, a
  unique or float column does not), and WRITE_BLOOM_FILTER is boolean-only (no
  per-column list), so there is nothing to add or "target" on the DuckDB write
  path. The one real gap is the arrow write paths (write_dataset for flat
  colData in BiocDuckDB::writeParquet, write_parquet for the empty / non-DuckDB
  coord block in DuckDBArray): R arrow (25.0.0) exposes no bloom-filter option
  at all, so those files carry no blooms. Options if this ever matters: route
  those writes through a DuckDB COPY (larger change; risks byte-parity with
  existing goldens), or accept it (blooms only help RARE-value equality lookups
  and do nothing for common values, so the practical cost is small). Revisit if
  R arrow gains bloom support.
