[kvlonge]

Why

DuckDB 2.0.0 was released today. ingestr installs the newest DuckDB ADBC driver when none is installed, so DuckDB files it wrote moved to the 2.0 storage format, which the locked duckdb 1.5.2 package (and SQLBuild's DuckDB adapter) cannot read. The ingestr DuckDB E2E tests started failing on every PR that runs them, and a user's ingestr-loaded DuckDB source could become unreadable by SQLBuild.

Changes

• Before running ingestr, SQLBuild installs the DuckDB ADBC driver matching its installed duckdb package into <environment>/etc/adbc/drivers with dbc, unless the user sets ADBC_DRIVER_PATH. A matching driver is reused; a failed install is a clear integration error naming the DuckDB version.
• The ingestr extra adds dbc and requires ingestr>=1.1.73 (lockfile updated from 1.0.0).
• New adbc_driver_install subprocess operation in the lifecycle catalogue and progress labels.
• Docs: the ingestr page explains the driver pinning and the ADBC_DRIVER_PATH override.

Verification

• Unit tests for the driver step: missing and mismatched drivers install the duckdb version, a matching driver and a user ADBC_DRIVER_PATH are left alone, a failed install names the version.
• The ingestr DuckDB E2E tests pass from a fresh home with no cached driver (they failed the same way as CI before the fix); sqlbuild-core tests pass.
• make check-ci.