fix: scd2 unmapped insert for nested tables

[deschman]

Description

For nested tables loaded with the scd2 merge strategy, the follow-up SQL that inserts new rows from the staging table into the destination used INSERT INTO … SELECT *. That ties column order to whatever the bulk writer produced in staging. On Spark/Databricks (ANSI mode), if staging and destination disagree on a column’s type—e.g. schema drift or evolving JSON—SELECT * can produce values that do not assign cleanly to the target table and surfaces as obscure DATATYPE_MISMATCH errors.

This change generates an explicit column list from the destination table schema (excluding the SCD2 from/to metadata columns) for both the INSERT target list and the SELECT, so values are aligned by column name with the destination DDL instead of by staging column order.

Related Issues

• Fixes scd2 nested table inserts are nondeterministic #3811

Additional Context

• Repro context and discussion: weak/nested schemas where child fields appear or disappear between loads; DuckDB often tolerates this with name-based parquet loads, while Databricks failed on the previous SELECT * pattern.
• I have read CONTRIBUTING.md.
• I ran make test-common locally and the suite was green.