Support auto_functionalized for None returns

[zou3519]

We support () returns but not None returns:

import torch

lib = torch.library.Library("mylib", "FRAGMENT")
# Change None to "()" and it will pass
lib.define("foo2(Tensor x, Tensor(a!) out) -> None")

def foo_impl(x, out):
    out.copy_(x)

lib.impl("foo", foo_impl, "CompositeExplicitAutograd")

x = torch.randn(3)
out = torch.zeros(3)

@torch.compile
def f(x, out):
    torch.ops.mylib.foo(x, out)

f(x, out)

cc @ezyang @msaroufim @bdhirsh @anijain2305 @chauhang