While investigating BF-32587, we have two predominant theories about what could be happening:

• The GuardedPageAllocator is somehow getting initialized twice, causing it to allocate memory, then forget that it allocated that memory when asked to deallocate it.
• Memory allocated by a seperate tcmalloc component is ending up in the address space that the GuardedPageAllocator thinks that it owns.

We can set up simple invariants to try to catch these situations, which would be a very useful debugging aid if triggered and otherwise should have a minimal impact on the system.