perf(duckdb): optimize generated SQL for ArrayConcat

[NickCrews]

Taking this ibis code:

import ibis

e = ibis.array([1]).concat([None], [2, 3], [4, 5, 6])
ibis.to_sql(e)

before:

SELECT
  CASE
    WHEN CASE
      WHEN CASE
        WHEN CAST([1] AS TINYINT[]) IS NULL OR [NULL] IS NULL
        THEN NULL
        ELSE LIST_CONCAT(CAST([1] AS TINYINT[]), [NULL])
      END IS NULL
      OR [2, 3] IS NULL
      THEN NULL
      ELSE LIST_CONCAT(
        CASE
          WHEN CAST([1] AS TINYINT[]) IS NULL OR [NULL] IS NULL
          THEN NULL
          ELSE LIST_CONCAT(CAST([1] AS TINYINT[]), [NULL])
        END,
        [2, 3]
      )
    END IS NULL
    OR [4, 5, 6] IS NULL
    THEN NULL
    ELSE LIST_CONCAT(
      CASE
        WHEN CASE
          WHEN CAST([1] AS TINYINT[]) IS NULL OR [NULL] IS NULL
          THEN NULL
          ELSE LIST_CONCAT(CAST([1] AS TINYINT[]), [NULL])
        END IS NULL
        OR [2, 3] IS NULL
        THEN NULL
        ELSE LIST_CONCAT(
          CASE
            WHEN CAST([1] AS TINYINT[]) IS NULL OR [NULL] IS NULL
            THEN NULL
            ELSE LIST_CONCAT(CAST([1] AS TINYINT[]), [NULL])
          END,
          [2, 3]
        )
      END,
      [4, 5, 6]
    )
  END AS "ArrayConcat((Array((1,)), (None,), (2, 3), (4, 5, 6)))"

after:

SELECT
  CASE
    WHEN CAST([1] AS TINYINT[]) IS NULL
    OR [NULL] IS NULL
    OR [2, 3] IS NULL
    OR [4, 5, 6] IS NULL
    THEN NULL
    ELSE LIST_CONCAT(LIST_CONCAT(LIST_CONCAT(CAST([1] AS TINYINT[]), [NULL]), [2, 3]), [4, 5, 6])
  END AS "ArrayConcat((Array((1,)), (None,), (2, 3), (4, 5, 6)))"

For a more real-world example, on the example function I give in #10996, before this change I get 8400 lines of SQL, afterwards it is 3400!

[cpcloud]

Nice find, thanks!