RFC: Cell::with_{mut_,}ref() should use finally

[bblum]

currently if the closure passed to with_ref/with_mut_ref fails, the cell will not put back the value that was taken from it. Is this desired? Or should it put it back?

[brson]

I think it should put the value back.

[bblum]

I'm inclined to agree. The possible argument against is analogous to the reason for poisoning in mutexARC and RWARC -- that failure somehow means the data is in a logically-invalid (though typechecker-valid) state. The idea of poisoning arcs has shrunk on me too, though.