[mypyc] Fix segfault in zip() with an unfinished native generator

[rheard]

Fixes mypyc/mypyc#1234.

A for loop over a call to a compiled generator function calls the generator's helper method directly (ForNativeGenerator, added in #19415). The helper sets the generator's return value only when the generator returns, so if the helper returns NULL without setting it, the generator raised. This was checked in gen_cleanup(), which runs when the loop exits. That works for a plain loop, but in zip() another iterator can end the loop while the generator is suspended. The return value was NULL then too, so the loop took its error path without an exception set, and crashed.

The check now happens in gen_condition(), in a block that's only reached when the generator stops, so ForNativeGenerator no longer needs a cleanup block. For other loops, the generated code is the same, just in a different order. No IR tests cover these loops, so none change.

[JukkaL]

Thanks for the fix!