Provide functions to sequence evaluations relative to io actions

[Borgvall]

This is my way to fix issue #1.

[treeowl]

I think this is a good idea, but I'm not convinced we need all three functions. On the flip side, we should probably also offer

stToEval :: (forall s. ST s a) -> Eval a

[Borgvall]

Technically one needs only runEvalIO. Then the other two functions can be easily defined using the public interface. I just provided them, to be symmetric with runEval, withStrategy and using. Typical usecases might look like this:

do   ...
     x <- expression `usingIO` strat
     y <- withStrategyIO strat2 . f . g $ x
     runEvalIO $ do
        ...

In my testings I never encountered a case, where withStrategyIO actually seems useful though. So it might can be droped.

stToEval might be useful. One might create a feature request with an usecase.

[simonmar]

This looks OK to me, I'll go ahead and merge. Thanks for adding the test!