testsuite: Perf tests that collect both run/alloc and ghc/alloc only compare ghc/alloc
Have a look at https://gitlab.haskell.org/ghc/ghc/-/jobs/1242614. I tried to reproduce the -1.4% run/alloc change in T21839r, a test which serves both as a run/alloc and ghc/alloc test. I failed to reproduce the change. In fact, I only measured 3.3MB of run/allocs locally, compared to 440MB (ish) in CI. That's strange!
Then I got an idea: I compared ghc/alloc and run/alloc measurements and it appears that both reported the same numbers. So it appears that ghc/alloc measurements overwrite run/alloc measurements in our testsuite. Worth sorting out, I think!