Python: Add support for global attribute writes

[tausbn]

I'm still not convinced that this is the best solution, but it fixes the
issue in the short term. When we eventually clean up the whole
EssaNode business, we can consider whether there's a better way to do
this.

---

I don't think this requires a change note, but will happily add one if needed.

[yoff]

Ideally, we would add a test showing the missed flow and then see the situation improve by the widened attribute write.

[tausbn]

Ideally, we would add a test showing the missed flow and then see the situation improve by the widened attribute write.

Ah, good point. I'll add one.

[yoff]

I found one of the tests confusing, otherwise this looks fine to me.

[yoff]

What are we testing here? That local_var is different from the one defined in the assignment test?
Perhaps add a comment to explain our expectations. Or perhaps assign to the variable inside the function?

[tausbn]

The intent was to show that even though we have an attribute assignment to a (local) variable that shadows a global of the same name, this does not result in additional flow. I can add global local_var if you think that'll make it clearer (though it is redundant). Assigning to the variable inside the function would make it a local variable, which would prevent the ModuleVariableNode from even existing.

[yoff]

Ah, the shadowing is an extra twist.

[yoff]

LGTM