New semantic analyzer: some assignments aren't recognized as types if incomplete

[JukkaL]

In the following test case the forward reference to In in the class body doesn't work:

[case testNewAnalyzerNamedTupleCallNested]
from typing import NamedTuple
o: C.Out
i: C.In
reveal_type(o)  # E: Revealed type is 'Tuple[Tuple[builtins.str, __main__.C.Other, fallback=__main__.C.In], __main__.C.Other, fallback=__main__.C.Out]'
reveal_type(o.x)  # E: Revealed type is 'Tuple[builtins.str, __main__.C.Other, fallback=__main__.C.In]'
reveal_type(o.y)  # E: Revealed type is '__main__.C.Other'
reveal_type(o.x.t)  # E: Revealed type is '__main__.C.Other'
reveal_type(i.t)  # E: Revealed type is '__main__.C.Other'

class C:
    Out = NamedTuple('Out', [('x', In), ('y', Other)])   # 'In' undefined  <<--- should be ok?
    In = NamedTuple('In', [('s', str), ('t', Other)])
    class Other: pass

The root cause is that if the rvalue of an assignment is incomplete, we add a PlaceholderNode that doesn't have becomes_typeinfo set. This may also affect other things such as assignment-based TypedDicts, type aliases, type variables and NewTypes (haven't checked).

Not sure what would be the best way to fix this. One idea would be to treat every placeholder as a possible type, at least if it's generated from an incomplete assignment. A second idea would be to recognize possible types by looking at an incomplete rvalue, but this doesn't seem always possible.

This is follow-up to #6983.