Update the test middleware for nREPL 0.6

[bbatsov]

nREPL 0.6 features a simplified and more robust evaluation model (see nrepl/nrepl#98), but this break the current implementation of the test middleware, which was making some assumptions about the internals of the eval middleware.

This ticket is a reminder to address this before nREPL 0.6 is released.

@jeffvalk would you like to tackle this? The updates should be pretty simple - seems to me that only the with-interruptible-eval macro should be updated.

[jeffvalk]

@bbatsov I'm not sure how soon I'd be able to get to this -- particularly if it's a research item.

[bbatsov]

I think it's just a matter of changing the use of queue-eval to the new exec as @cgrand did in the PR where he reworked the eval thread management.