New analyser: better processing of import cycles with star imports

Consider this test case:

[case testClassDeferredStar]
import lib
[file lib.py]
from other import *

class C:
    x: int
[file other.py]
import lib

Currently the order of processing targets is like this:

I think it should be possible to process such SCC without deferral at all. Couple ideas to achieve this:

Better ordering of targets within SCC
Make missing_names act per namespace (to avoid extra deferral in more complex scenarios)

Currently this affects two plugin test cases testTypeInAttrDeferredStar and testTypeInDataclassDeferredStar.

It looks like we generally process modules in a very sub-optimal order. I'll see what happens if we reverse the order.

Fixed by #7163.

Navigation Menu