I guess, Python technically counts, although the type annotations were one hell of an afterthought.

I do feel like the two features are somewhat at odds with each other.
A REPL is useful for quickly trying something out, which you’d typically use to put together a quick script. In a larger application, you generally want to define unit tests, which make your “quickly trying something out” reproducible and automateable.
Meanwhile, type annotations primarily start to shine in larger applications, where you cannot keep the whole context in your head and where refactorings become problematic, when nothing checks that the parts still work together.

In some sense, they also don’t work well together. Type annotations work best when the whole context is known. But in a REPL, you almost always have an incomplete context, because you’re still typing everything out.
So, it cannot type-check the first few lines you write and it becomes awkward to try to show type errors once the context is complete, because the code isn’t on-screen anymore.