feat(adbc_insert): pass driver-specific statement options

[jatorre]

adbc_insert only sets adbc.ingest.target_table and adbc.ingest.mode on the statement, so there's no way to reach ingest options that drivers expose at statement level.

The motivating case is the Snowflake driver: its GeoArrow auto-detect correctly maps planar-edge input (what DuckDB produces) to GEOMETRY, and expects callers who want GEOGRAPHY to say so with adbc.snowflake.statement.ingest_geo_type (see adbc-drivers/snowflake#167). Other drivers have similar statement-level knobs, such as ADBC 1.1's adbc.ingest.temporary, adbc.ingest.target_catalog and adbc.ingest.target_db_schema.

This adds options :=, a STRUCT or MAP with the same shape adbc_connect accepts. It is applied after the target table and mode. Keys adbc_insert already owns (adbc.ingest.target_table, adbc.ingest.mode) are rejected, and unknown keys surface the driver's own error.

SELECT * FROM adbc_insert(conn, 'places', (SELECT * FROM src), mode := 'create',
  options := {'adbc.snowflake.statement.ingest_geo_type': 'geography'});

Tests (test/sql/adbc_insert.test, SQLite driver) use adbc.ingest.temporary to show the option reaches the driver: the table lands in temp, not main. They cover both STRUCT and MAP forms and the three error paths (reserved key, non-STRUCT/MAP value, unknown driver option). The full test/sql/* suite passes locally with HAS_ADBC_SQLITE_DRIVER=1 (834 assertions; the Postgres/MySQL/Trino server tests were skipped).

[rustyconover]

Hi @jatorre looking at this now.

[rustyconover]

Looks good to me, I'll merge once CI goes green! Thanks @jatorre

[rustyconover]

@jatorre waiting for merge into community extensions repo: duckdb/community-extensions#2776

[jatorre]

that was quick! thanks!