fix(std_common): handle NUM_READY_WORKER correctly

[Koichi98]

Description

The following three points were fixed in kernel/src/main.rs because the shell would not start when executing std_common with make run-std:

1. num_cpu() is only available after each CPU has called increment_num_cpu(). If we don't wait for all CPUs to increment, NUM_READY_WORKER underflows and the following while loop never terminates.

while NUM_READY_WORKER.load(Ordering::SeqCst) > 0 {
    awkernel_lib::delay::wait_microsec(10);
}

I added wait_microsec(100) before num_cpu() to avoid this. Ideally, we'd reference the CPU count obtained in each kernel/src/arch/ARCH/kernel_main.rs, but since this would require significant changes, I've decided not to include it in this PR.

2. Non-primary CPUs should only decrement NUM_READY_WORKER after PRIMARY_READY becomes true in std-common, as it is in other architectures. Therefore, I modified std_common to also check that PRIMARY_READY is TRUE before decrementing.

3. Replaced fetch_sub with fetch_update to prevent underflow when num_cpu is less than the actual number of CPU. Added a panic when NUM_READY_WORKER can't be decremented (already 0) as this indicates an unexpected state.

Related links

How was this PR tested?

Verified that the shell correctly enters the wait state by executing 'make run-std RELEASE=1' more than 10 times

Notes for reviewers