New semantic analyzer: Clean-up logic for classes in functions

[ilevkivskyi]

While working on #6421 I looked again at logic for storing symbols for local (function- or method-level) classes in prepare_class_def(). The logic is too ad-hoc and needs to be cleaned up. Here are some points that should be fixed:

• defn.fullname and info.fullname() should always coincide
• info.fullname() prefix should coincide with the symbol table where the info is actually stored (for serialization)
• things should work for double nested classes
• things should work in fine-grained mode

I think the logic should be as following. Consider this example:

class Out:
    def meth(self) -> None:
        class In:
            def meth(self) -> None:
                class C: pass

The class C should be stored in its local namespace under name 'C'. In addition it should be stored in the namespace of __main__.Out under name 'C@5'. info.fullname() and defn.fullname should be both '__main__.Out.C@5'.

[MatthieuDartiailh]

What is the current status of this ? I saw that a couple of PRs addressed part of the issue but #7281 is not fixed and I am wondering how much of the initial comment still applies after three PR related to this issue. I added typing to a project but this issue will prevent downstream package to benefit of it so I may be interested in helping.

[MatthieuDartiailh]

Also how should the system behaves for a class defined in a function ? Should the symbol be stored on the function or at the global level ?