There is no obvious logic in wrapping or not wrapping new attributes into versioned comments:

MariaDB [test]> create table t1 (pk int, c text compressed invisible);

Query OK, 0 rows affected (0.18 sec)

MariaDB [test]> show create table t1 \G

*************************** 1. row ***************************

       Table: t1

Create Table: CREATE TABLE `t1` (

  `pk` int(11) DEFAULT NULL,

  `c` text /*!100301 COMPRESSED*/ INVISIBLE DEFAULT NULL

) ENGINE=InnoDB DEFAULT CHARSET=latin1

1 row in set (0.00 sec)

Here COMPRESSED is wrapped (wrongly btw, since it was only pushed in 10.3.2), but INVISIBLE is not at all.

I have a plausible explanation for it: COMPRESSED is optional, because it doesn't change the application logic (much); but INVISIBLE might, so maybe it's better that it fails on servers which don't support invisible columns. However, I don't know if it was really meant this way.