Go back to backup/restore SpinLockIRQ system

[roblabla]

See #190

This is not the final solution. API is still unsound, but it at least works when the invariants are upheld...

[todo]

Find sane design for SpinLockIRQ safety

Currently, SpinLockIRQ API is unsound. If the guards are dropped in the wrong order, it may cause IF to be reset too early.
Ideally, we would need a way to prevent the guard variable to be reassigned. AKA: prevent moving. Note that this is different from what the Pin API solves. The Pin API is about locking a variable in one memory location, but its binding may still be moved and dropped. Unfortunately, Rust does not have a way to express that a value cannot be reassigned.
Another possibility would be to switch to a callback API. This would solve the problem, but the scheduler would be unable to consume such locks. Maybe we could have an unsafe "scheduler_relock" function that may only be called from the scheduler?

---

https://github.com/roblabla42/KFS/blob/8eed485f3b8085f9df4fdb27d1c10e869e6d76cd/kernel/src/sync.rs#L53-L63

---

This comment was generated by todo based on a TODO comment in 8eed485 in #191. cc @roblabla.

[marysaka]

lgtm

[Orycterope]

Nice