Refactor parial types to avoid code duplication

Currently there is some duplication around partial types:

• Logic in try_infer_partial_generic_type_from_assignment() essentially duplicates the logic inlined in check_assignment() for None partial types. These two probably can be refactored to the same method if we also update handle_partial_var_type() to not special-case partial None types, and instead consistently return a partial type in lvalue context.
• Logic in try_infer_partial_type_from_indexed_assignment() duplicates that in try_infer_partial_type() (there is already a TODO item). This one can be refactored by either pushing the latter a bit down the call stack (closer to check_call()), or by generating a synthetic CallExpr with __setitem__ and passing it to try_infer_partial_type().

The second item may be something to watch out when implementing support for these:

a = defaultdict(list)
a[0].append('yes')

b = {}
b.setdefault(0, []).append('yes')

c = defaultdict(set)
c[0].add('yes')

d = {}
d.setdefault(0, set()).add('yes')

@JukkaL this is probably not something important, but maybe it makes sense to fix this while we are at it?

Start by locating try_infer_partial_generic_type_from_assignment(), check_assignment(), handle_partial_var_type(), try_infer_partial_type_from_indexed_assignment(), and try_infer_partial_type(). Compare their partial-type handling, then evaluate the defaultdict and setdefault examples as regression cases; done means the duplicated logic is unified without breaking those inference scenarios.