[FEATURE]: Build Asynchronous TCP Connection Pooling Engine with Pre-Warmed Fallback Descriptors

Problem

When a Layer 4 proxy triggers a connection hot-swap under less than 2 milliseconds, creating a brand-new TCP socket handshake on-demand introduces transient synchronization spikes. This can choke the line-rate transport pipeline during high-concurrency multi-tenant traffic storms.

Proposed Solution

Scaffold a high-throughput, non-blocking connection pool manager inside internal/proxy/pool_manager.go:

1. The Pre-Warmed Standby Ring: Implement background routines (goroutines) that continually spin up and maintain an active array of pre-warmed, authenticated socket descriptors connected directly to fallback destination nodes natively.
2. Atomic Descriptor Handover: Utilize lock-free pointer atomic swaps (sync/atomic) to immediately drop the failing connection socket and splice the active stream onto an already open standby socket file descriptor without running full connection setups midway.

Use Case

Guarantees strict clamping of failover latency bounds to absolute sub-millisecond flat, making FaultPlane completely indestructible under heavy transactional loads.

Start by reading internal/proxy/pool_manager.go and the surrounding proxy and transport code to understand where connection creation and failover are handled. Define the pool lifecycle, standby authentication, atomic handover behavior, and measurable latency and cleanup requirements before implementing; done requires a working non-blocking fallback path under concurrent failover.