head.hackage should use a version of GHC with assertions enabled

In ghc#22521 (closed) we have an example of a change which causes an assertion failure when trying to compile head hackage. This was not caught by CI because, as Ryan points out in that ticket, head.hackage uses a release GHC instead of one with assertions enabled.

I think we should run head.hackage using a compiler with assertions enabled. Were we to also run benchmarks, those would still have to be with a released compiler, in order to avoid potential performance regressions.