In the workspace, there are planning and budget constraints. For example, much of Java's Date api is deprecated (because it was horrible), but that's something that's hard to replace gradually. And I can't just replace it in the whole massive codebase without getting budget approval, so it doesn't get fixed for years.

In my hobby projects, I tend to fix all warnings including deprecations almost immediately. But few of those projects ever get finished. Perfect is the enemy of done...

Yes, if it makes my job easier today. I prefer to fix that later, once the feature is removed.

That being said, this is risky, and should only do if you have a mature level of control of your dependencies (you're using pinned dependencies, you are not automatically increasing to the next major version).

Also, I'd do this for software I have full control of (an API, and internal service, etc...) and not for something that I'd give to my customers, like a library.