Fix the logic in active self-type calculation for current scope

[ilevkivskyi]

Fixes #5846

The fix for crash is kind of straightforward. Previously active_self_type() always returned None at method scope (sic!)

While working on this I discovered couple other problems:

• Area around checking LSP is generally problematic (see also Overriding definition such as type alias in a subclass causes a crash #5425 that I tried to fix earlier but failed). In particular, the are two ways to get current TypeInfo, one is from lvalue.node.info or defn.info, another is using active_self_type() (this is rather abusing it IMO). I don't try to fix this here (i.e. switch to always using one way) because this is a relatively large refactoring.
• Currently in type checker we defer nested functions instead of top-level ones. I believe this is not by design and is rather caused by absence of docstrings and unintuitive method names in CheckerScope class. Namely, there "top function" stays for "top of stack function", not "top-level function". I don't try to fix this here either because this is conceptually a big change.

[JukkaL]

I agree that it's strange that this hasn't been causing more trouble. I assume that this only triggers in very specific cases that are pretty rare.

[JukkaL]

What if the type is incompatible with the base class?

[ilevkivskyi]

Hm, I am not super happy with the error message, because it says:

Incompatible types in assignment (expression has type "int", base class "B" defined the type as "str")

while the initial type is T (it shows the mapped/expanded type). It is however not easy to improve, so I think it is OK.