DuckDB engine as build-tagged artifact + Parquet/CSV/Iceberg querying

go-duckdb requires CGO — conflicts with the single-static-binary promise, so: core binary stays pure Go; DuckDB ships as a build-tagged second artifact (dbui-duckdb) with per-OS release builds. The wedge is the file story: query Parquet/CSV/Iceberg from the same client that speaks PG/MySQL. Dialect spec + acceptance suite pass = done ("DuckDB-first" framing dropped — harlequin owns it).

**Acceptance:** dbui-duckdb artifact in the release matrix; acceptance suite green on the DuckDB dialect; querying a local Parquet file demoed.

Start with the release matrix, build-tag setup, and DuckDB dialect acceptance suite, then trace how existing database artifacts are built. Done means the dbui-duckdb artifact is included for each OS, the DuckDB dialect acceptance suite passes, and a local Parquet query is demonstrated.

Written by the indexing model from the issue text.