os/kernel/semaphore: fix timeout+post race in sem_timedwait

[rish-sg]

Fix a race condition between timeout and sem_post that causes spurious wakeups on subsequent semaphore waits.

Root Cause:
When timeout fires, sem_waitirq() increments semcount to restore the count that was decremented when the thread blocked. If a concurrent sem_post() also increments semcount before the thread gets CPU, there will be a leftover semcount (double increment).

This leftover causes the next sem_wait() on the same semaphore to return immediately without actually waiting (spurious wakeup).

Fix:
In the error path after sem_wait() returns, check if:

• errno is ETIMEDOUT (timeout fired)
• semcount > 0 (concurrent post happened)

If both conditions are true, consume the leftover semcount and return OK since the semaphore was successfully acquired.

This is POSIX-compliant per:
"Under no circumstance shall the function fail with a timeout if the semaphore can be locked immediately."