Lock-free event distribution for informerProcessor

[radiohead]

Problem

Phase 2 removed the RWMutex from the hot path in informerProcessor.distribute(), but the processor still uses mutex-based coordination internally. Further performance gains are possible with lock-free data structures.

Current bottlenecks:

• Mutex contention during event distribution
• Allocation overhead in steady state

Proposed Solution

Replace the mutex-based informerProcessor with an xsync MPMC (Multi-Producer Multi-Consumer) lock-free queue.

Key features:

• Zero allocations in steady state
• Lock-free event enqueue/dequeue
• Worker pool with configurable parallelism
• Blocking enqueue semantics (prevents event loss)

Expected Impact

Based on prototyping:

• 6.9x throughput improvement: 350K → 2.4M events/sec
• Zero allocations in steady state: 896B/op → 0B/op
• CPU contention reduction: 58.63% → <5%
• Latency: 2847ns → 412ns per operation

Dependencies

• Go module: github.com/puzpuzpuz/xsync/v3
• Opt-in via InformerOptions.UseLockFreeProcessor

Related

• Epic: Improve operator informer performance
• Design doc: .claude/docs/performance/lock-free-processor-implementation.md
• Research: .claude/docs/performance/lock-free-queues-research.md