Alloc access tracking shows spurious reads after each write

[RalfJung]

After each write we do validation which causes reads, and these reads actually trigger the before_memory_read machine hook and then cause alloc access tracking to print.

It seems to me like we probably want to not call the hook for such "fake" reads at all? These reads don't actually exist in the AM; usually the value would get validated "while copying" but Miri doesn't support that so we do some extra reads.