2. A missing method definition is assumed to be bottom. (I don't like this rule. Ghc with suitable options warns in such case.

This definitely sounds like a bad idea... seems to violate intuition about what an instance of a class is... if you can't define a method of the class, you probably don't belong in that class.

The first point is important. It's possible that an overloaded function in the next version of a library is promoted to a method with a default definition, and then existing code using that class doesn't break. Understood. Not sure what I think of it. Wouldn't it be better for the new library to include a new class deriving from the old? That would more clearly indicate the relationship... if the class itself changes, the instances should change too.