RISCV32 smp support is broken

[d3zd3z]

CI tests on RISCV32 smp targest suggests synchronization/locking issues. The two issues seen are:

• A deadlock on the first call to printk, suggesting an incompatibility with with the critical section on the rust side, and the spin locks on the Zephyr side. This is a little surprising, since the critical section should be using spin locks directly.
• Multi-thread tests have failures that suggest conflicting access to shared data, which also suggests that the critical section implementation (or the SpinMutex implementation) isn't working.

[d3zd3z]

I have fixed the use of spinlocks for critical sections, and switched to irq_lock()/irq_unlock(). However, I still seem to be seeing failures on this. I'm actually getting now what appears to be a corruption that is happening the first time k_thread_resume is being called from Rust. There are various interactions between spinlocks and the irq_lock, so something is probably happening there.