This optimization does not apply to Postgres in general, but only to citrus_fdw and other column-store foreign data wrappers.
The Postgres execution engine was designed to operate on a row-store. Even so, it can operate on a column-store via a foreign data wrapper like citrus_fdw. However, to do so it must currently reconstruct rows-at-a-time to fit its row-oriented nature. As a result, it cannot currently realize the potential performance benefits of using a column-store for certain queries.
This article is about an extension that adds some new column-at-a-time aggregate functions, then hooks into the Postgres executor to modify eligible query plans to use them. It thus enables Postgres to take advantage of the column-oriented nature of the citrus_fdw.
Is that an accurate way to look at it?