Treat methods on protocols with empty bodies as abstract

[JarnoRFB]

When declaring a protocol it is suggested in the documentation that method that have to be defined in implementations of that protocol can be marked with ... and I find this syntax quite pretty. However, currently the ... is treated as a default implementation. Which can lead to confusing behavior. An example is the following:

from typing_extensions import Protocol


class SupportsAdding(Protocol):
    def add(self, x: int, y: int) -> int: ...


class Adder(SupportsAdding):
    def subtract(self, x: int, y: int) -> int:
        return x + y


if __name__ == '__main__':
    adder: SupportsAdding = Adder()
    z = adder.add(1, 2)
    print(type(z))
    reveal_type(z)

Given the rather drastic misspelling of add as subtract, mypy still infers the protocol to be correctly implemented. The result of reveal_type(z) is int, while it is actually NoneType. PEP 544 recommends to decorate non-default implementations with @abstractmethod. However, this leads to less elegant code and additional runtime checks. Since it seems kind of obvious that ... is never a default implementation of a method, I would suggest to treat it the same way as protocol methods decorated with @abstractmethod.

I am using mypy version 0.761.

[ilevkivskyi]

This is essentially a particular case of #2350

(Btw I wanted to fix this at some point but decided it is not worth it)