Judge loss by arrival order and a deadline, one tx ring per stream

docs/measurement.md

@@ -69,7 +69,7 @@ Measured on the bench host (i9-13900H, 6P+8E, 20 threads), RAPL package power at
 Measured in the E810 era; the binding constraint was total CPU across ~28 goroutines on 20 threads.
 
 - **Splitting TX senders from RX streams** — raises TX but collapses RX (RX scales with queue count, capped at 7 on that NIC).
-- **CPU pinning** — the Go scheduler beats manual placement; E-cores are poor at TX.
+- **CPU pinning for performance** — the Go scheduler beats manual placement; E-cores are poor at TX. (The TX feeders are nevertheless each locked to a dedicated cpu, spaced off shared hyperthreads: XPS keys the hardware ring on the sending cpu, and a stable per-stream ring is an ordering requirement, not a speed one.)
 - **Batch sizes above 64** — no gain, worse loss.
 - **Chasing the TX `sendmmsg` busy-spin on ENOBUFS** — it only exists when the TX ring is full, which means the wire is the ceiling; recovering that CPU buys no packets. POLLOUT is inert under PACKET_QDISC_BYPASS (skb freed on ENOBUFS, socket always reports writable).
 - **Letting idle cores sleep by relaxing cpu_dma_latency** (200 µs admits the 127 µs C2_ACPI) — steady-state deep-idle residency ~2% and power unchanged: the Go scheduler spreads wakeups across every thread, so no core idles long enough for the idle governor to pick a deep state.