thor: Implement reentrant logging from NMI handler

[managarm]

thor: Implement reentrant logging from NMI handler

[avdgrinten]

This can be implemented by:

1. Adding a per-CPU ring buffer to buffer the logs from that CPU. A kernel thread should be added to dump the per-CPU ring buffers to a global ring buffer.

2. Adding a per-CPU atomic variable that acts as a lock to detect whether we are are currently writing to the per-CPU ring buffer. In case that we are currently writing while an NMI happens, a UAR-like mechanism can be used to re-write the partially written data to the per-CPU ring buffer. Afterwards, the NMI can safely perform its own logging.

A later improvement could drop the per-CPU buffers in favor of a lock-free global ring buffer (while keeping the UAR-like mechanism) but it would be reasonable to implement the simpler mechanism using per-CPU buffers first.