Support multiple output files

[simonmar]

We had "groups" as a way to run a test multiple ways where the output files might differ, but that only works if you can run the test multiple ways without recompiling. For tests where the output depends on something in the environment, we need a way to manage multiple output files, so that's what this does.

Additionally I've used it in the Haskell indexer tests, where the GHC version affects the output.