If we panic in guest due to running out of memory, allocating inside the panich andler will cause another panic, causing infinite panic recursion until the stack is overflown and we hit a guard page, which exits as a Stackoverflow error.

Our current tests for running out of memory tries to allocate a huge chunk which fails, but the allocation inside the panic handler succeeds because it's way smaller than the first one... :/