Introduce slots; remove mutable boxes/vectors/record fields/object fields

[pcwalton]

IIRC we agreed upon the notion of a "slot", which is a simple way to create a mutable interior value. This lets us abstract over the mutable and immutable vectors without the subtyping relation that "mutable?" introduces, which is painful (and which we never got right in the presence of generics).

So you could construct a slot using, say, slot 3, and that creates a mutable pseudo-box that is interior. Slots have no representation at the IR level; they're purely a way to help us enforce immutability restrictions. [mutable int] would become [slot int], @mutable int would become @slot int, etc.

This may be appropriate for the first milestone, depending on how serious we consider the type soundness hole with generics to be for 0.1.

[marijnh]

So taking the value of a slot in a non-lvalue context would be an error?

let x = slot int;
f(x);

[pcwalton]

Actually thinking through this more, slots have several weird issues relating to autoderef and generic functions. Talked this over with Dave and I think it makes sense to keep mutable?, but to fix the issue with covariance/invariance and rename it to something a little less gross (const maybe?)

[pcwalton]

Dupe of issue #518.