`re_rebalance_coherence` doesn't properly account for fundamental types

[nikomatsakis]

As part of an effort to stabilize the re_rebalance_coherence RFC (#63599), I was creating tests trying to get more coverage of all possible scenarios, and I found that we don't seem to be handling fundamental types in the way that the RFC specifies.

In particular this test is presently accepted and yet ought to be rejected:

#![feature(re_rebalance_coherence)]

extern crate coherence_lib as lib;
use lib::*;
use std::rc::Rc;

struct Local;

impl<T> Remote1<Local> for Box<T> { /* should error */ }
impl<T> Remote1<Local> for &T { /* should error */ }

fn main() {}

I'll be adding this test (along with some others) in a PR soon with "FIXME" comments.