Report IPE information of thunks involved in cycles

Currently debugging cyclic thunks (e.g. the NonTermination exception, which manifests as the dreaded <<loop>> message) is quite difficult. I generally have to resort to ghc-debug or gdb to understand such cases. However, with IPE information we now have the ability to determine the provenance of thunks. It would be nice if GHC could attach the set of IPE information (and perhaps cost-centers, if profiling) involved in the cycle to the NonTermination exception.