Like the warmup of AffinitySupport, we better do AllocationProfilingSupport warmup early, so that sanityCheck_Footprints that would run soon would hit the warm code. This saves about 10 ms per forked VM.