Regression: The behaviour of `rparWith r0` has changed

[Borgvall]

For years the documented (and actual) behaviour of rparWith r0, was to create a spark that does no evaluation. However since commit 9ea4c07 the spark will evaluate to weak head normal form.

I have restored the documented behaviour in pull request #34.

[treeowl]

No, it never had that behavior.

[Borgvall]

It did. Just run the test program with ghc 8.0.2 and parallel 3.2.1.0.

[treeowl]

I see. I was confused by the documentation suggesting rparWith s = rpar `dot` s. The documentation definitely needs to be clarified. Tonight I'll come up with a fix that doesn't use realWorld#.