[mypyc] Fix regression with nested functions

[ilevkivskyi]

Fixes #16480

Fix is straightforward, but also suspicious, since I am not sure how it ever worked without this.

[mr-c]

I can confirm that this fixes my original issue, thank you!

[ilevkivskyi]

Some more info for reviewers (@JukkaL @msullivan @ichard26): currently visit_symbol_node() is only called from visit_name_expr(), but the former method has two branches: for when we already seen the name, and when we see it first time. These branches are hit when a NameExpr appears as a reference, or as an assignment l.h.s. (i.e. definition), correspondingly. This is why free_variables are set correctly for variables, but for functions this was broken, because visit_symbol_node() was not called on function definition, only when it appears as reference.

[ichard26]

I can't merge, but LGTM. I appreciate the explanation, it was helpful to understand the patch!