It's a bit difficult statically because it is a flow sensitive analysis.
You are not wrong that it is a sharp edge. Completely removing nils from interfaces is not possible because:
1. not backward compatible
However I would nuance a little. Having an empty interface ie. a untyped nil is useful.
Having typed nils in interfaces is arguable.
Because every value type that has methods can make pointer. That means potential deref if any such pointer is passed to an interface variable instead of the value itself.
Being able to keep nil from some interfaces would be useful.
You're not wrong.
In general there is not much value in having working methods on a typed nil pointer.
If we think in terms of bottom wrt type theory, yes it is supposed to implement every type. But that would be closer to untyped nil and that's not how go's type system works either. It is close though. We just don't have a language concept for nillable int because variables are auto initialized to 0. And because it would be difficult to encode such an information purely virtually.
But that could be possible in theory, without mechanical sympathy. I digress. The takeaway is that I don't think a linter can do the trick easily but there has been good attempts.
And it is worth pondering, you're right.
