Fast-servers pushes leaner high-throughput server architecture

Strong systems thinking here: it is less about new primitives and more about using `epoll`/`kqueue` with stricter architecture choices.

• –Frames mainstream worker-pool patterns as legacy bottlenecks for modern multicore machines.
• –Gives concrete implementation guidance (thread affinity, accept loop, socket options) instead of abstract advice.
• –Useful to backend and infra engineers optimizing latency and throughput, even outside AI workloads.
• –Not AI-specific, but relevant as foundational performance engineering for developer infrastructure.