Use memory_order_acquire for spinlock instead of acq_rel

[HFTrader]

Summary

• Use memory_order_acquire for lock() and try_lock() instead of memory_order_acq_rel
• The standard spinlock pattern is acquire on lock, release on unlock. The release semantics on the lock path are unnecessary — when acquiring a lock, you need to ensure subsequent operations stay after the acquisition (acquire), but there is nothing to release at that point
• This restores the original behavior from the initial commit (e57758e, Jan 2014). The change to acq_rel was introduced in 8530b8a (Dec 2016) with the rationale "since test-and-set is both a read and a write operation, it should [use acq_rel]" — however the "write" in test-and-set is the lock flag itself, which does not need release semantics during acquisition

Impact

• x86: No change. test_and_set compiles to XCHG which is a full barrier regardless of the memory order specified
• ARM/POWER: Removes an unnecessary barrier instruction. acq_rel generates both an acquire and release barrier, while acquire generates only the acquire barrier. This is a measurable improvement on weakly-ordered architectures

Context

This aligns with the three open PRs improving the spinlock (#147, #149, #158) and the discussion in issue #146.

Test plan

• All existing unit tests pass
• Consistent with the acquire/release pattern used by the existing unlock() method