Refactor deriving code between auto_encode.rs and deriving.rs

[nikomatsakis]

Both auto_encode and deriving.rs do similar things and share various similar routines. We should probably write a "deriving utility library" that they can all use. Deriving already seems to be factored in this way, so maybe it's just a matter of making auto_encode use the routines found there.

[nikomatsakis]

I left a few FIXMEs scattered about.

[huonw]

Is there a particular reason why the traits in auto_encode shouldn't be factored to be normal derivable traits? e.g.

#[deriving(Eq, Clone, Encodable, Decodable)]
struct Foo { bar: int }

[nikomatsakis]

@dbaupp no, they should be. I think @erickt had started some work in that direction, actually.

[alexcrichton]

Closed as part of #6851