A fish is unable to leave, your point is merely that it's "not easy" to leave. If you're on Gitlab then it seems pretty easy to just download your code and make an account somewhere else.
As surprising as it may sound, I think more fish manage to wiggle away from the hook than are caught. Ask any fisherman. They wiggle away with a mouth full of blood, which is what the users who have to migrate from a lock-in situation figuratively do.
Let's turn the argument on its head: GitLab can still offer the features for free, it just chooses not to. And unlike the users they have full control over the situation. GitLab decided to pull the trigger and the users have to dodge and wiggle to get away with "only" superficial wounds. That's a shitty way to deal with the situation. They could have very well grandfathered the accounts with the freebies but instead they took advantage of the situation that people have a harder time leaving than they had coming, and the power imbalance this creates.
For people on Gitlab, your point about "downloading the code and moving on" really only applies to the simplest point of use on Gitlab. Anybody using Gitlab's CI/CD pipelines has everything in their repo(s) fully setup only for that, and not for other CI/CD offerings. Moving their stuff is often much much much more work than "just download your code", and sometimes with time/money constraints, completely infeasible in the short term.
Your comment brushes aside this dynamic, in favor of a naive model where users don't succumb to path dependence or local maxima. Why?