Nightly jobs run with -dcore-lint which seriously affects quality of allocations numbers

Cross-post from head.hackage#42 (closed)

Compiling with -dcore-lint skews allocation numbers significantly, to the extent that we are really just benchmarking the speed of core lint rather than anything relevant to the users. (See #20825)

When I build Cabal locally with HEAD I get 161,723,938,240 bytes allocated but grafana reports > 600GB!, which matches more closely to a number I get when running with -dcore-lint.