allow canceling semantics?

[jdegoes]

The idea is something like this:

newtype Aff e a = Aff ((Error -> Eff e Unit) -> (a -> Eff e Unit) -> EffA e (Unit -> Aff e Unit)

So an Aff can decide to allow canceling. The semantics would be something along the lines of:

1. Calling the cancel function does not guarantee canceling.
2. If the cancel function succeeds, the Aff will error, not succeed.
3. Cancelations in derived Aff would be propagated.

The main advantage is that IO can be aborted, e.g. the <|> combinator for Par can abort the other computation, which can lead to efficiency gains.

The public API would be something like:

killAff :: forall e a. Error -> Aff e a -> Aff e Boolean

Which returns a true if it's killed and a false otherwise. Or something like that.