[mypyc] Use correct rtype for variable with inferred optional type

[ichard26]

def f(b: bool) -> None:
    if b:
        y = 1
    else:
        y = None

f(False)

On encountering y = 1, mypyc uses the expression type to determine the
variable rtype. This usually works (as mypy infers the variable type
based on the first assignment and doesn't let it change afterwards),
except in this situation.

NameExpr(y)'s type is int, but the variable should really be int | None.
Fortunately, we can use the Var node's type information which is
correct... instead of blindly assuming the first assignment's type is
right.

Fixes mypyc/mypyc#964

[twoertwein]

Is there a reason to limit combining infered types across branches to None? For example, pyright consistently combines any types from different branches:

def test(flag: bool):
    if flag:
        x = 1
    else:
        x = ""
    reveal_type(x)
    # mypy: builtins.int
    # pyright: Literal[1, ""]

I assume this might be a big change for mypy but when adding a special case for None, it seems only consistent to use the same logic more broadly.

[ichard26]

@twoertwein that is a question for the mypy maintainers. I'm part of the core team, but I only work on mypyc. This PR simply fixes mypyc (the compiler) to do the right thing when mypy infers a combined type across branches (and not generate code that immediately crashes). In what situations mypy infers a combined type is not within the domain of this PR.

Edit: found the relevant mypy issue -> #6233.

[JukkaL]

Thanks for the fix! Looks good. Left some suggestions about tests.

[JukkaL]

Looks good now, thanks!