in one project I was working on, we used a similar version of the 2nd method from the article:

Every table had the same table with _del suffix (eg. users_del). If a record was deleted, it was simply moved to _del table. We used code for this but later we started to use db triggers.

It worked quite well, and yes, there was always someone who wanted to undelete things. One downside was, if the schema changed on the source table, we needed to also change the schema in _del table. I like the approach with storing the data as json. That way there could be only 1 deleted_stuff table because it was looking quite strange having all the _del tables.