Mooncake issue #4209: TCP transport first transfer after an idle gap fails with queue-full rejections; PD decode hangs until timeout

Shows the failure mode of falling back to TCP when RDMA is unavailable. 'the first KV transfer between two engine instances succeeds, but a transfer submitted after a short idle gap (~5 s) fails': every slice of the batch is rejected by the TCP lane queue (queue-full rejection count 1 -> 2 -> 4 -> ... -> 512+ within milliseconds), batch_transfer_sync_write returns -1, and the error is relayed to the consumer. The consumer then sits in WAITING_FOR_REMOTE_KVS and the client hangs until its own timeout; a request sent ~90 s later succeeds. Net effect: 'deployments observe either 60-90 s TTFT spikes after every idle period or outright request failures'. Env: mooncake-transfer-engine 0.3.13.post1 and 0.3.14rc1, vLLM 0.29.0 built-in MooncakeConnector, 1P1D on one host, 2x H20 96GB, mooncake_protocol tcp with MC_FORCE_TCP=1, Qwen3-8B.