Test failures because instances are loaded in a different order depending on optimisations

Some test cases, such as T7627 and ghci011, have an output that depends on optimisations, which can cause spurious test failures. It would be nice to normalise the output of these tests by sorting the output to avoid this issue.