thor: Report OOM to user space about managarm

[avdgrinten]

This can work as follows:

• Attach a "policy group" to each thread. The policy group should determine whether threads can still allocate when the system is near its physical memory limit. That way, servers are protected from resource exhaustion while allocations by normal programs can fail (their allocations are "failable").
• Once the physical memory usage exceeds some high watermark, let failable allocations fail. In particular, allocations that are on-demand should result in SIGBUS while non-on-demand allocations should fail in helAllocateMemory.
• Optionally: add a Failable flag to helAllocateMemory such that servers can still perform failable memory allocations.