add MPMC queue (Vyukov design)

[AneesPatel]

What

MPMC ring buffer using per-slot sequence counters. Based on Dmitry Vyukov's classic design - each slot coordinates its own producer/consumer handoff instead of a single shared lock.

Why this vs the existing SPSC buffer

The existing LockFreeRingBuffer only works with one producer and one consumer. This one handles arbitrary N producers and M consumers safely. The tradeoff is slightly more overhead per operation (one extra atomic load per slot).

Notes

• Head/tail live on separate cache lines (alignas(64)) to avoid false sharing - worth it when multiple threads hammer both ends.
• Capacity must be a power of two.
• Push/pop are non-blocking and return false/nullopt if full/empty.
• Stress tested: 4 producers x 4 consumers, 1M items, verify the sum matches.

Files

• include/MPMCQueue.h
• tests/test_mpmcqueue.cpp
• CMakeLists.txt