APPROVE. Neat fix for #6694: explicit column list in the jsonb_populate_recordset INSERT so Postgres applies column DEFAULTs (and NOT NULL DEFAULT columns like served_at TIMESTAMP DEFAULT now() NOT NULL) instead of SELECT * forcing NULL into every omitted column.

• CDC path also lists the __feldera_op / __feldera_ts metadata columns explicitly — consistent with the materialized path and with the existing pipeline invariant that value_schema.fields doesn't already contain those.
• Case-sensitive columns stay quoted (test materialized_insert_quotes_case_sensitive_columns locks that in).
• Integration test creates a real NOT NULL DEFAULT now() column outside the Feldera schema and asserts every row's served_at was populated from the DEFAULT — the exact failure mode from #6694.

One optional follow-up (non-blocking): the CDC branch pushes cdc_op_column/cdc_ts_column into insert_columns while the materialized branch does not — if a future refactor merges those config fields into value_schema, the CDC branch will silently double-list them. A one-line debug_assert that insert_columns has no duplicates would catch that regression cheaply, but it's not worth blocking on now.