perf: replace seq_cst with acquire-release memory ordering - #10

Changes

• Replace default std::memory_order_seq_cst operations with explicit memory ordering.
• Use acquire/release synchronization between producer and consumer.
• Keep local index reads as relaxed operations where synchronization is unnecessary.
• Update benchmark implementation for more accurate throughput measurement.
• Unified MSVC runtime library configuration to /MD
• Enabled gtest_force_shared_crt for Windows builds

CMakeLists Motivation

Windows CI failed during linking because GoogleTest and project targets used different MSVC runtime libraries:

• Project: /MD (Dynamic CRT)
• GoogleTest: /MT (Static CRT)

This caused LNK2038 and multiple definition errors.

Observation

The performance improvement compared with seq_cst is limited.

This suggests that the current bottleneck is likely not memory ordering itself, but cache-line contention between write_idx and read_idx.

The next optimization will focus on cache-line alignment and false sharing elimination.