Updated queued_spinlock to latest Linux implementation

[tiagormk]

Ported changes from the queued lock implementation in Linux kernel v6.6.7 in addition to general improvements.

General improvements:

• Change CONFIG_NR_CPUS to the max allowed without affecting the lock bit layout.
• Fix allocation of mcs_pool. Used cache-aligned allocations to make sure there is no false-sharing between cpus (in the actual kernel implementation DEFINE_PER_CPU_ALIGNED is used). Cache line size is assumed to be 64 bytes.
• Removed intermediate struct __qspinlock definition.

Lock behavior changes from Linux v6.6.7:

• Limit number of spinning iteration in the initial lock->pending check.
• lock->pending set using fetch_or instead of an cmpxchg loop.
• Backup spin if cannot allocate MCS node (should not happen in lockhammer).
• Reduce the number of cmpxchg operations performed when acquiring the lock after being released form the MCS node
• Added missing atomic primitives to lk_atomics.h