Change "mutable?" to something else ("const"?) and fix covariance/invariance

[pcwalton]

AIUI the general consensus was that we should implement a mutable "slot" type. This is basically equivalent to an interior record with one mutable field. This allows us to remove the mutable? botch.

[graydon]

Assigning to first release so that it gets cleaned up before anyone relies on it in current state.

[pcwalton]

Actually I'm not sure about this; I can't figure out what the deref semantics need to be. I think mutable? may be the best solution...

[pcwalton]

See issue #856. We should at least fix covariance/invariance for mutable?.

[brson]

The variance of mutable? is fixed by 0f0fa26

[brson]

With the variance issue fixed can we punt on mutable? for 0.1?

[brson]

mutable? is now const