E.g. in
```rust
match x {
1 => function1,
2 => function2,
}
```
we need to try coercing both to pointers. Turns out this is a special case in
rustc as well (see the link in the comment).
Fixes a lot of false type mismatches.
(And as always when touching the unification code, I have to say I'm looking
forward to replacing it by Chalk's...)
It improves compile time in `--release` mode quite a bit, it doesn't
really slow things down and, conceptually, it seems closer to what we
want the physical architecture to look like (we don't want to
monomorphise EVERYTHING in a single leaf crate).
