Aliasing hints are based on AST types, but should be based on monormorphized ty::t

[nikomatsakis]

We have some simple hints to detect when ~T pointers are passed and tell LLVM they are unique. These hints should be based on the monomorphized ty::t, but they are currently based on the AST. This isn't wrong, it just misses optimization opportunities.

See FIXME.

[msullivan]

What optimization opportunities does this miss?

[thestinger]

If you have a generic function where T expands to ~int. It would be easy to fix.

[msullivan]

Yeah, I realized that just as I pressed comment. This should be super easy to fix. I can do it.