Supporting (or not) invalid identifiers in **kwargs

I am tired of litigating this, and doubly so if the litigation is based on documentation that was written decades ago. (These docs are also clearly out of date, positional-only arguments are still described as a CPython-only feature for builtins only.)

I never intended this to be an undefined language feature (those are generally a bad idea). Can someone please submit a doc PR that clarifies the situation, so I can approve it?