Just now I briefly looked at implementations of Maybe in Java and at least one of them fakes this with checked exceptions, while another seems to lean heavily on Guava's Function interface, which is just awkward to use. I'd certainly be interested in seeing alternative implementations, if you know of any that are decent.
Also, a maybe more serious issue is that since it's not idiomatic Java you'd have to do a lot of wrapping around libraries, and hard-selling to colleagues. Admittedly no longer dealing with flaws in Java's type system, but a big part of a language is the community and ecosystem, and in Haskell's case people using it have already bought on to the advantages of stronger types.
You're quite right that programming with these kinds of types in Java introduces you to a whole new kind of Hell! I'm not sure I'd recommend it.
However, I'm simply pointing out that types like these are _not_ language features. Some languages may be better suited to them than others, but they are definitely library features, and had best not be considered otherwise.
No, because your Maybe or Either reference may still be null, so you get "Maybe fuck you" and "Either fuck you, or fuck you". That's the first part.
The second part is, due to the lack of match completeness check (or more generally match) and the shitty type system, APIs forcing developers to safely unwrap or rewrap values are... shitty.
fuckYou :: a
fuckYou = fuckYou
justTrolling = Just fuckYou
That said, however, I'm very sympathetic to the rejection of non-total languages. But we must note that even though Haskell is inconsistent as a logic, it is still able to derive tolerable advantage from the use of option types, etc.