Since you're making heavy use of generics there, I suspect that what you're seeing is a bridge method problem where the runtime invocation may come in as a compiler-generated bridge method instead of the originally declared method. The bridge method won't have any annotations on it, so method validation would obviously fail then.
However, unfortunately, I'm getting mixed results even with an explicit call to our BridgeMethodResolver. Maybe it is picking up the method from the sub-interface as opposed to the base interface and only finding the constraint annotations in one but not the other case. A JVM is quite free to play games with reflection here.
This turns out to be rather involved, and indeed only occuring for generically typed methods. We'll certainly do something about it (if Hibernate Validator itself doesn't, since it could resolve such an ambiguous Method handle as well), but probably only in the 4.x line.
I cannot reproduce this against Hibernate Validator 5.x on JDK 8 anymore, going through the Bean Validation 1.1 method validation API. So I assume this has been addressed on Hibernate Validator's side in the meantime.