Conflicts:
	src/runtime/builtin_modules/builtins.cpp
	src/runtime/builtin_modules/sys.cpp

Conflicts:
	src/runtime/code.cpp

added BoxedCode::traverse and BoxedCode::dealloc

Conflicts:
	from_cpython/Python/traceback.c
	src/codegen/unwinding.cpp
	src/runtime/builtin_modules/sys.cpp
	src/runtime/capi.cpp
	src/runtime/cxx_unwind.cpp
	src/runtime/exceptions.cpp

Conflicts:
	src/capi/modsupport.cpp

made some refcounting changes but did not change the cffi patch

I had copied in their refcounting code around it but our data
format wasn't the same.  whoops.

It returns an object that keeps the returned array alive.

I think this is more than is necessary but it's so hard to reason about that.

And then I saw that we don't rewrite delattr anyway.

Sometimes to pass a variable number of argument we allocate an array, store
the arguments in that array, and then pass the array as the single C-level
arg.  This caused ref issues where the refcounter thought that the variables
were dead after the store -- ie before the function call -- and it would
decref them right then.  This commit adds a "refUsed" hook that allows us
to specify these sorts of "non-IR-based" uses.

Another option would have been to make the arrays fully refcounted (ie the
array itself is treated as owning references to its contents), but in this
case I think that's overkill since it's really just an ABI issue that the
refcounter didn't understand.