Unify MonadIO and MonadThrow constraints?

[snoyberg]

A number of functions provide both MonadIO and MonadThrow constraints. However, by sticking a liftIO in front of the function body, we can knock this down to just a MonadIO constraint in all cases. Is there any objection to this change? If not, I can send a PR.

[mrkkrp]

I think this is actually a good idea, please go ahead with the PR. I wonder if the same can be done about MonadCatch. I'd start by moving ignoringIOErrors (internal helper) into plain IO and then adding liftIO in front of copyDirRecurGen and similar, the constraint should go away then, although maybe I'm missing something.

We could try to stop depending on exceptions directly then. path still depends on exceptions though.

[snoyberg]

PR #28 is open.