And so engineers implement new features and write more code and release the features and get rewarded for doing so. Once the feature is done, if it's easy to delete it they will. If not, and it never is, they'll leave it around and move on to something else.
When you have 3 billion users if you make the company 1/100th of a penny per user per year you pay for yourself. That's like increasing the value of the average user by 1/200,000. So your feature doesn't even have to be that good. It just has to be minimally better than what it's replacing. Restyle a button to be more appealing in the peripheral vision of users and that probably breaches that threshold.
Reality is that it isn't quite this easy (fewer users are that valuable, most don't use it daily, etc). But it gets the point across. You work there to make more money and deleting some feature that is probably referenced by some other dead code somewhere else isn't profitable work for anybody.