[Feat/Infra]: Implement Stateful TCP Connection Handover Mechanisms during Live Failover Transitions

Problem

When an upstream model instance or inference server crashes, traditional transport layers drop the active socket descriptor. This forces the client application runtime to spin up a manual reconnect loop, leading to dropped connection frames and transaction payload leakage mid-flight.

Proposed Solution

Build a stateful connection tracking migration routine within internal/gateway/server.go.

• Capture and duplicate state bit arrays of active file descriptors (FD) immediately upon ingress.
• Implement an upstream socket hot-swapping mechanism that safely transitions an active stream from a degraded socket to a healthy fallback cluster node using a persistent TCP connection pool interface.
• Maintain synchronization of payload metadata registries so the fallback receiver node can decode from the exact byte boundary failure index.

Alternatives

• Forcing hard client-side retries, which shifts the resilience configuration complexity onto the application developers and triggers token processing overhead.

Use Case

Guarantees uninterrupted processing for long-running workflows under heavy infrastructure degradation, ensuring failure masking with zero context data loss.

Additional Notes

This must integrate with the binary formats developed under issue #5 and maintain compatibility with multi-node replication registries.

Start by reading internal/gateway/server.go and the binary formats referenced in issue #5. Trace the existing ingress and connection-handling entry points before assessing the proposed state migration and fallback behavior. Done means active TCP streams can transition to a healthy cluster node while preserving payload metadata, byte-boundary state, and multi-node replication compatibility.