Yes, we could choose to make the addition of cases to an enum an ABI-breaking change instead of enacting Jordan's proposal. But that, too, is a "change to enums." If we make no changes to enums, then adding cases to an enum would result in undefined behavior, because quite literally we will not have defined what the behavior should be in that circumstance.
A protocol requirement is a protocol requirement is a protocol requirement. Even with ABI stability, new entries can be added to protocol witness tables, period.
And my purpose it is to convince you that you ought to change your mind and to view it as a desirable state of affairs!
There is a fundamental difference between protocols and classes, and it can be illustrated pretty clearly as follows. Consider a module whose source code is the following, in its entirety:
public class A { /* concrete implementations */ }
public protocol B { /* requirements */ }
extension B { /* default implementations */ }
Class A is a useful type. You can't inherit from it, but it can do something. In today's Swift, protocol B is a useful protocol. You can't instantiate it, but it can be conformed to. In your vision for future Swift, protocol B can be entirely deleted from the module's compiled binary; it's all wasted code. Now, that seems like an undesirable state of affairs.