RFC: Handle built-in type names in resolve, rather than parser

[marijnh]

This was propsed by @brson in #1586 . The parser doesn't need to know about built-in type names like int, f64, and bool. We could create an artificial top-level scope in the resolver that resolves these to the right types. It'd simplify the parser and the AST.

This might also we worth doing for simple value keywords like true, false, maybe even log, which don't influence the syntax around them. I already did it for self recently, and that definitely resulted in cleaner code.

Open question is whether shadowing of these would be allowed. I'd be okay with either, but lean slightly towards allowing it, since that makes them behave exactly like other names and removes a special case / error check from the compiler.

[graydon]

Smells like a case for an optional warning/error (lint like).

Personally I'd still ban them, but I know I'm minority opinion on this and am ok being overruled by everyone else's wrong, permissive taste.

[nikomatsakis]

I'm ok with it. One thing I have long wanted to do is to rename the ctypes module to c and then change types like c_int to be written c::int. Not sure if this is actually a good idea, but this change would make it possible. =)

[pcwalton]

Yeah, I'm ok with this.