[mypyc] Merge yield_from_except_op

[TH3CHARLie]

relates mypyc/mypyc#753

This PR merges yield_from_except_op into the new IR, including several changes:

• A change to C wrapper's signature since the new IR currently has no unified way to represent the address of a pointer.
• A change to LoadAddress, allowing it to load local reg's address.
• A change to uninit pass, suppressing checks on LoadAddress

[JukkaL]

Thanks -- this is a tricky one! I have some thoughts about potentially making this cleaner. It's kind of awkward to have to use casts from PyObject * to PyObject **.

[JukkaL]

Why do we need this? Add comment explaining this, since this is a bit surprising. I wonder if we can somehow get rid of this special case.

[TH3CHARLie]

We need this because we are taking the address of an uninitialized object, the automatically inserted initialization check will produce incorrect code, leading the control flow directly to uninitialized error rather than the following load address and other ops.

[JukkaL]

Can you check that there is some existing case that uses the primitive? If not, it makes sense to add one (but I assume that this is tested somewhere).

[TH3CHARLie]

Can you check that there is some existing case that uses the primitive? If not, it makes sense to add one (but I assume that this is tested somewhere).

yes, there's testYield for this(but I think it's not completely for this op itself but also so other yield and exception stuff)

[TH3CHARLie]

I've made some updates according to your review, including using object_pointer_rprimitive and a new IR test. Unfortunately, the special case in the uninit pass is somewhat inevitable

[JukkaL]

Thanks for the updates! Looks good.

I'd suggest dropping the test case (but even then it has been useful to see what sort of code is being generated).