[red-knot] Attribute assignments in @staticmethods should not influence instance attributes

[sharkdp]

The goal of this issue is to get rid of the TODO comments in this test. An attribute assignment in a @staticmethod should not influence the existence of implicit (instance) attributes:

class Unrelated:
    x: int

class C:
    @staticmethod
    def f(unrelated: Unrelated):
        unrelated.x = 1

reveal_type(C().x)  # should be an error

Whether or not this requires special-casing of @staticmethod during semantic index building is an open question. The answer to this question influences whether or not we can also handle the more exotic edge-cases in the mentioned test (e.g. aliased @staticmethod).

Part of: #14164